Fix inference layout and remove packaging files

.gitignore

@@ -1,34 +0,0 @@
-# Python
-__pycache__/
-*.py[cod]
-.pytest_cache/
-*.egg-info/
-
-# Virtual environment
-.venv/
-venv/
-
-# IDE
-.vscode/
-.idea/
-
-# Environment and credentials
-.env
-*.token
-
-# Hugging Face cache
-.hf_cache/
-.cache/
-huggingface/
-
-# Large/local data
-data/
-models/
-checkpoints/
-outputs/
-results/
-logs/
-*.log
-
-# OS
-.DS_Store

configuration_mistral.py

@@ -0,0 +1,53 @@
+# coding=utf-8
+"""Minimal local configuration for the GenomeOcean custom Mistral model."""
+
+from transformers.configuration_utils import PretrainedConfig
+
+
+class MistralConfig(PretrainedConfig):
+    model_type = "mistral"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=32000,
+        hidden_size=4096,
+        intermediate_size=14336,
+        num_hidden_layers=32,
+        num_attention_heads=32,
+        num_key_value_heads=8,
+        hidden_act="silu",
+        max_position_embeddings=4096 * 32,
+        initializer_range=0.02,
+        rms_norm_eps=1e-6,
+        use_cache=True,
+        pad_token_id=None,
+        bos_token_id=1,
+        eos_token_id=2,
+        tie_word_embeddings=False,
+        rope_theta=10000.0,
+        sliding_window=4096,
+        attention_dropout=0.0,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.sliding_window = sliding_window
+        self.num_key_value_heads = num_key_value_heads or num_attention_heads
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.attention_dropout = attention_dropout
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )

ensemble_manifest.json

@@ -2,7 +2,6 @@
   "folds": [
     {
       "fold": "fold1",
-      "source": "/mnt/taskmaster1/scratch/hyejong/01_gv_genomeocean/finetuning_go_sub_add/ft_models/train_100M_v1.2_5kb/fold1/checkpoint-44240",
       "files": [
         "config.json",
         "configuration_mistral.py",
@@ -15,7 +14,6 @@
     },
     {
       "fold": "fold2",
-      "source": "/mnt/taskmaster1/scratch/hyejong/01_gv_genomeocean/finetuning_go_sub_add/ft_models/train_100M_v1.2_5kb/fold2/checkpoint-48620",
       "files": [
         "config.json",
         "configuration_mistral.py",
@@ -28,7 +26,6 @@
     },
     {
       "fold": "fold3",
-      "source": "/mnt/taskmaster1/scratch/hyejong/01_gv_genomeocean/finetuning_go_sub_add/ft_models/train_100M_v1.2_5kb/fold3/checkpoint-46530",
       "files": [
         "config.json",
         "configuration_mistral.py",
@@ -41,7 +38,6 @@
     },
     {
       "fold": "fold4",
-      "source": "/mnt/taskmaster1/scratch/hyejong/01_gv_genomeocean/finetuning_go_sub_add/ft_models/train_100M_v1.2_5kb/fold4/checkpoint-44195",
       "files": [
         "config.json",
         "configuration_mistral.py",
@@ -54,7 +50,6 @@
     },
     {
       "fold": "fold5",
-      "source": "/mnt/taskmaster1/scratch/hyejong/01_gv_genomeocean/finetuning_go_sub_add/ft_models/train_100M_v1.2_5kb/fold5/checkpoint-46130",
       "files": [
         "config.json",
         "configuration_mistral.py",

examples/example_ncldv_mirus.fna

@@ -1,26 +0,0 @@
->synthetic_sub_example 5000 clean bases; N separators are removed by preprocessing
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC
-CCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCCNCCCCCCCCCC

fold1/config.json

@@ -16,9 +16,17 @@
   "eos_token_id": null,
   "hidden_act": "silu",
   "hidden_size": 768,
+  "id2label": {
+    "0": "NCLDV",
+    "1": "Mirus"
+  },
   "initializer_range": 0.02,
   "intermediate_size": 3072,
   "is_causal": true,
+  "label2id": {
+    "Mirus": 1,
+    "NCLDV": 0
+  },
   "max_position_embeddings": 32768,
   "model_type": "mistral",
   "num_attention_heads": 8,

fold2/config.json

@@ -16,9 +16,17 @@
   "eos_token_id": null,
   "hidden_act": "silu",
   "hidden_size": 768,
+  "id2label": {
+    "0": "NCLDV",
+    "1": "Mirus"
+  },
   "initializer_range": 0.02,
   "intermediate_size": 3072,
   "is_causal": true,
+  "label2id": {
+    "Mirus": 1,
+    "NCLDV": 0
+  },
   "max_position_embeddings": 32768,
   "model_type": "mistral",
   "num_attention_heads": 8,

fold3/config.json

@@ -16,9 +16,17 @@
   "eos_token_id": null,
   "hidden_act": "silu",
   "hidden_size": 768,
+  "id2label": {
+    "0": "NCLDV",
+    "1": "Mirus"
+  },
   "initializer_range": 0.02,
   "intermediate_size": 3072,
   "is_causal": true,
+  "label2id": {
+    "Mirus": 1,
+    "NCLDV": 0
+  },
   "max_position_embeddings": 32768,
   "model_type": "mistral",
   "num_attention_heads": 8,

fold4/config.json

@@ -16,9 +16,17 @@
   "eos_token_id": null,
   "hidden_act": "silu",
   "hidden_size": 768,
+  "id2label": {
+    "0": "NCLDV",
+    "1": "Mirus"
+  },
   "initializer_range": 0.02,
   "intermediate_size": 3072,
   "is_causal": true,
+  "label2id": {
+    "Mirus": 1,
+    "NCLDV": 0
+  },
   "max_position_embeddings": 32768,
   "model_type": "mistral",
   "num_attention_heads": 8,

fold5/config.json

@@ -16,9 +16,17 @@
   "eos_token_id": null,
   "hidden_act": "silu",
   "hidden_size": 768,
+  "id2label": {
+    "0": "NCLDV",
+    "1": "Mirus"
+  },
   "initializer_range": 0.02,
   "intermediate_size": 3072,
   "is_causal": true,
+  "label2id": {
+    "Mirus": 1,
+    "NCLDV": 0
+  },
   "max_position_embeddings": 32768,
   "model_type": "mistral",
   "num_attention_heads": 8,

modeling_mistral.py

@@ -0,0 +1,1617 @@
+# coding=utf-8
+# Copyright 2023 Mistral AI and the HuggingFace Inc. team. All rights reserved.
+#
+# This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
+# and OPT implementations in this library. It has been modified from its
+# original forms to accommodate minor architectural differences compared
+# to GPT-NeoX and OPT used by the Meta AI team that trained the model.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+""" PyTorch Mistral model."""
+import inspect
+import math
+import warnings
+from typing import List, Optional, Tuple, Union
+from dataclasses import dataclass
+
+import torch
+import torch.nn.functional as F
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
+
+from transformers.activations import ACT2FN
+from transformers.cache_utils import Cache, DynamicCache
+from transformers.modeling_attn_mask_utils import (
+    _prepare_4d_causal_attention_mask, 
+    _prepare_4d_causal_attention_mask_for_sdpa, 
+    _prepare_4d_attention_mask,
+    _prepare_4d_attention_mask_for_sdpa,
+)
+from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast, SequenceClassifierOutputWithPast, ModelOutput
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import (
+    add_start_docstrings,
+    add_start_docstrings_to_model_forward,
+    is_flash_attn_2_available,
+    is_flash_attn_greater_or_equal_2_10,
+    logging,
+    replace_return_docstrings,
+)
+from .configuration_mistral import MistralConfig
+
+
+if is_flash_attn_2_available():
+    from flash_attn import flash_attn_func, flash_attn_varlen_func
+    from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input  # noqa
+
+    _flash_supports_window_size = "window_size" in list(inspect.signature(flash_attn_func).parameters)
+    print("Using flast_attn 2")
+    print(f"flash_attn_func supports window_size: {_flash_supports_window_size}")
+
+
+
+logger = logging.get_logger(__name__)
+
+_CONFIG_FOR_DOC = "MistralConfig"
+
+
+# Copied from transformers.models.llama.modeling_llama._get_unpad_data
+def _get_unpad_data(attention_mask):
+    seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
+    indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
+    max_seqlen_in_batch = seqlens_in_batch.max().item()
+    cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.int32), (1, 0))
+    return (
+        indices,
+        cu_seqlens,
+        max_seqlen_in_batch,
+    )
+
+
+# Copied from transformers.models.llama.modeling_llama.LlamaRMSNorm with Llama->Mistral
+class MistralRMSNorm(nn.Module):
+    def __init__(self, hidden_size, eps=1e-6):
+        """
+        MistralRMSNorm is equivalent to T5LayerNorm
+        """
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(hidden_size))
+        self.variance_epsilon = eps
+
+    def forward(self, hidden_states):
+        input_dtype = hidden_states.dtype
+        hidden_states = hidden_states.to(torch.float32)
+        variance = hidden_states.pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
+        return self.weight * hidden_states.to(input_dtype)
+
+
+# copied from transformers.models.llama.modeling_llama.LlamaRotaryEmbedding with Llama->Mistral
+# TODO @Arthur no longer copied from LLama after static cache
+class MistralRotaryEmbedding(nn.Module):
+    def __init__(self, dim, max_position_embeddings=2048, base=10000, device=None):
+        super().__init__()
+
+        self.dim = dim
+        self.max_position_embeddings = max_position_embeddings
+        self.base = base
+        inv_freq = 1.0 / (self.base ** (torch.arange(0, self.dim, 2, dtype=torch.int64).float().to(device) / self.dim))
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+
+        # Build here to make `torch.jit.trace` work.
+        self._set_cos_sin_cache(
+            seq_len=max_position_embeddings, device=self.inv_freq.device, dtype=torch.get_default_dtype()
+        )
+
+    def _set_cos_sin_cache(self, seq_len, device, dtype):
+        self.max_seq_len_cached = seq_len
+        t = torch.arange(self.max_seq_len_cached, device=device, dtype=torch.int64).type_as(self.inv_freq)
+
+        freqs = torch.outer(t, self.inv_freq)
+        # Different from paper, but it uses a different permutation in order to obtain the same calculation
+        emb = torch.cat((freqs, freqs), dim=-1)
+        self.register_buffer("cos_cached", emb.cos().to(dtype), persistent=False)
+        self.register_buffer("sin_cached", emb.sin().to(dtype), persistent=False)
+
+    def forward(self, x, seq_len=None):
+        # x: [bs, num_attention_heads, seq_len, head_size]
+        if seq_len > self.max_seq_len_cached:
+            self._set_cos_sin_cache(seq_len=seq_len, device=x.device, dtype=x.dtype)
+
+        return (
+            self.cos_cached[:seq_len].to(dtype=x.dtype),
+            self.sin_cached[:seq_len].to(dtype=x.dtype),
+        )
+
+
+# Copied from transformers.models.llama.modeling_llama.rotate_half
+def rotate_half(x):
+    """Rotates half the hidden dims of the input."""
+    x1 = x[..., : x.shape[-1] // 2]
+    x2 = x[..., x.shape[-1] // 2 :]
+    return torch.cat((-x2, x1), dim=-1)
+
+
+# copied from transformers.models.llama.modeling_llama.apply_rotary_pos_emb
+# TODO @Arthur no longer copied from LLama after static cache
+def apply_rotary_pos_emb(q, k, cos, sin, position_ids, unsqueeze_dim=1):
+    """Applies Rotary Position Embedding to the query and key tensors.
+
+    Args:
+        q (`torch.Tensor`): The query tensor.
+        k (`torch.Tensor`): The key tensor.
+        cos (`torch.Tensor`): The cosine part of the rotary embedding.
+        sin (`torch.Tensor`): The sine part of the rotary embedding.
+        position_ids (`torch.Tensor`):
+            The position indices of the tokens corresponding to the query and key tensors. For example, this can be
+            used to pass offsetted position ids when working with a KV-cache.
+        unsqueeze_dim (`int`, *optional*, defaults to 1):
+            The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
+            sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
+            that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
+            k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
+            cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
+            the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
+    Returns:
+        `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
+    """
+    cos = cos[position_ids].unsqueeze(unsqueeze_dim)
+    sin = sin[position_ids].unsqueeze(unsqueeze_dim)
+    q_embed = (q * cos) + (rotate_half(q) * sin)
+    k_embed = (k * cos) + (rotate_half(k) * sin)
+    return q_embed, k_embed
+
+
+class MistralMLP(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.intermediate_size = config.intermediate_size
+        self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
+        self.act_fn = ACT2FN[config.hidden_act]
+
+    def forward(self, x):
+        return self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
+
+
+# Copied from transformers.models.llama.modeling_llama.repeat_kv
+def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
+    """
+    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
+    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
+    """
+    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
+    if n_rep == 1:
+        return hidden_states
+    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim)
+    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
+
+
+class MistralAttention(nn.Module):
+    """
+    Multi-headed attention from 'Attention Is All You Need' paper. Modified to use sliding window attention: Longformer
+    and "Generating Long Sequences with Sparse Transformers".
+    """
+
+    def __init__(self, config: MistralConfig, layer_idx: Optional[int] = None):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        if layer_idx is None:
+            logger.warning_once(
+                f"Instantiating {self.__class__.__name__} without passing a `layer_idx` is not recommended and will "
+                "lead to errors during the forward call if caching is used. Please make sure to provide a `layer_idx` "
+                "when creating this class."
+            )
+
+        self.hidden_size = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_dim = self.hidden_size // self.num_heads
+        self.num_key_value_heads = config.num_key_value_heads
+        self.num_key_value_groups = self.num_heads // self.num_key_value_heads
+        self.max_position_embeddings = config.max_position_embeddings
+        self.rope_theta = config.rope_theta
+        self.is_causal = config.is_causal
+        self.attention_dropout = config.attention_dropout
+
+        if (self.head_dim * self.num_heads) != self.hidden_size:
+            raise ValueError(
+                f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}"
+                f" and `num_heads`: {self.num_heads})."
+            )
+        self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=False)
+        self.k_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=False)
+        self.v_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=False)
+        self.o_proj = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=False)
+
+        self.rotary_emb = MistralRotaryEmbedding(
+            self.head_dim,
+            max_position_embeddings=self.max_position_embeddings,
+            base=self.rope_theta,
+        )
+
+    def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
+        return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Cache] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+        **kwargs,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        if "padding_mask" in kwargs:
+            warnings.warn(
+                "Passing `padding_mask` is deprecated and will be removed in v4.37. Please make sure use `attention_mask` instead.`"
+            )
+        bsz, q_len, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states)
+        value_states = self.v_proj(hidden_states)
+
+        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+        value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            if self.layer_idx is None:
+                raise ValueError(
+                    f"The cache structure has changed since version v4.36. If you are using {self.__class__.__name__} "
+                    "for auto-regressive decoding with k/v caching, please make sure to initialize the attention class "
+                    "with a layer index."
+                )
+            kv_seq_len += past_key_value.get_seq_length(self.layer_idx)
+        cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
+
+        if past_key_value is not None:
+            cache_kwargs = {"sin": sin, "cos": cos}  # Specific to RoPE models
+            key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+        # repeat k/v heads if n_kv_heads < n_heads
+        key_states = repeat_kv(key_states, self.num_key_value_groups)
+        value_states = repeat_kv(value_states, self.num_key_value_groups)
+
+        attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt(self.head_dim)
+
+        if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len):
+            raise ValueError(
+                f"Attention weights should be of size {(bsz, self.num_heads, q_len, kv_seq_len)}, but is"
+                f" {attn_weights.size()}"
+            )
+
+        if attention_mask is not None:
+            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
+                raise ValueError(
+                    f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
+                )
+
+            attn_weights = attn_weights + attention_mask
+
+        # upcast attention to fp32
+        attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype)
+        attn_weights = nn.functional.dropout(attn_weights, p=self.attention_dropout, training=self.training)
+        attn_output = torch.matmul(attn_weights, value_states)
+
+        if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
+            raise ValueError(
+                f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is"
+                f" {attn_output.size()}"
+            )
+
+        attn_output = attn_output.transpose(1, 2).contiguous()
+        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
+
+        attn_output = self.o_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights, past_key_value
+
+
+class MistralFlashAttention2(MistralAttention):
+    """
+    Mistral flash attention module. This module inherits from `MistralAttention` as the weights of the module stays
+    untouched. The only required change would be on the forward pass where it needs to correctly call the public API of
+    flash attention and deal with padding tokens in case the input contains any of them.
+    """
+
+    # Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2.__init__
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+        # TODO: Should be removed once Flash Attention for RoCm is bumped to 2.1.
+        # flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignement, that was made default for flash_attn>=2.1. This attribute is used to handle this difference. Reference: https://github.com/Dao-AILab/flash-attention/releases/tag/v2.1.0.
+        # Beware that with flash_attn<2.1, using q_seqlen != k_seqlen (except for the case q_seqlen == 1) produces a wrong mask (top-left).
+        self._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal_2_10()
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Cache] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+        **kwargs,
+    ):
+        if "padding_mask" in kwargs:
+            warnings.warn(
+                "Passing `padding_mask` is deprecated and will be removed in v4.37. Please make sure use `attention_mask` instead.`"
+            )
+
+            # overwrite attention_mask with padding_mask
+            attention_mask = kwargs.pop("padding_mask")
+        bsz, q_len, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states)
+        value_states = self.v_proj(hidden_states)
+
+        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+        value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            if self.layer_idx is None:
+                raise ValueError(
+                    f"The cache structure has changed since version v4.36. If you are using {self.__class__.__name__} "
+                    "for auto-regressive decoding with k/v caching, please make sure to initialize the attention class "
+                    "with a layer index."
+                )
+            kv_seq_len += past_key_value.get_seq_length(self.layer_idx)
+
+        # Because the input can be padded, the absolute sequence length depends on the max position id.
+        rotary_seq_len = max(kv_seq_len, position_ids[:, -1].max().item()) + 1
+        cos, sin = self.rotary_emb(value_states, seq_len=rotary_seq_len)
+
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
+
+        use_sliding_windows = (
+            _flash_supports_window_size
+            and getattr(self.config, "sliding_window", None) is not None
+            and kv_seq_len > self.config.sliding_window
+        )
+
+        if not _flash_supports_window_size:
+            logger.warning_once(
+                "The current flash attention version does not support sliding window attention, for a more memory efficient implementation"
+                " make sure to upgrade flash-attn library."
+            )
+
+        if past_key_value is not None:
+            # Activate slicing cache only if the config has a value `sliding_windows` attribute
+            cache_has_contents = past_key_value.get_seq_length(self.layer_idx) > 0
+            if (
+                getattr(self.config, "sliding_window", None) is not None
+                and kv_seq_len > self.config.sliding_window
+                and cache_has_contents
+            ):
+                slicing_tokens = 1 - self.config.sliding_window
+
+                past_key = past_key_value[self.layer_idx][0]
+                past_value = past_key_value[self.layer_idx][1]
+
+                past_key = past_key[:, :, slicing_tokens:, :].contiguous()
+                past_value = past_value[:, :, slicing_tokens:, :].contiguous()
+
+                if past_key.shape[-2] != self.config.sliding_window - 1:
+                    raise ValueError(
+                        f"past key must have a shape of (`batch_size, num_heads, self.config.sliding_window-1, head_dim`), got"
+                        f" {past_key.shape}"
+                    )
+
+                if attention_mask is not None:
+                    attention_mask = attention_mask[:, slicing_tokens:]
+                    attention_mask = torch.cat([attention_mask, torch.ones_like(attention_mask[:, -1:])], dim=-1)
+
+            cache_kwargs = {"sin": sin, "cos": cos}  # Specific to RoPE models
+            key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+        # repeat k/v heads if n_kv_heads < n_heads
+        key_states = repeat_kv(key_states, self.num_key_value_groups)
+        value_states = repeat_kv(value_states, self.num_key_value_groups)
+        dropout_rate = 0.0 if not self.training else self.attention_dropout
+
+        # In PEFT, usually we cast the layer norms in float32 for training stability reasons
+        # therefore the input hidden states gets silently casted in float32. Hence, we need
+        # cast them back in float16 just to be sure everything works as expected.
+        input_dtype = query_states.dtype
+        if input_dtype == torch.float32:
+            if torch.is_autocast_enabled():
+                target_dtype = torch.get_autocast_gpu_dtype()
+            # Handle the case where the model is quantized
+            elif hasattr(self.config, "_pre_quantization_dtype"):
+                target_dtype = self.config._pre_quantization_dtype
+            else:
+                target_dtype = self.q_proj.weight.dtype
+
+            logger.warning_once(
+                f"The input hidden states seems to be silently casted in float32, this might be related to"
+                f" the fact you have upcasted embedding or layer norm layers in float32. We will cast back the input in"
+                f" {target_dtype}."
+            )
+
+            query_states = query_states.to(target_dtype)
+            key_states = key_states.to(target_dtype)
+            value_states = value_states.to(target_dtype)
+
+        # Reashape to the expected shape for Flash Attention
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
+        value_states = value_states.transpose(1, 2)
+
+        attn_output = self._flash_attention_forward(
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            q_len,
+            dropout=dropout_rate,
+            use_sliding_windows=use_sliding_windows,
+        )
+
+        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size).contiguous()
+        attn_output = self.o_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights, past_key_value
+
+    def _flash_attention_forward(
+        self,
+        query_states,
+        key_states,
+        value_states,
+        attention_mask,
+        query_length,
+        dropout=0.0,
+        softmax_scale=None,
+        use_sliding_windows=False,
+    ):
+        """
+        Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token
+        first unpad the input, then computes the attention scores and pad the final attention scores.
+
+        Args:
+            query_states (`torch.Tensor`):
+                Input query states to be passed to Flash Attention API
+            key_states (`torch.Tensor`):
+                Input key states to be passed to Flash Attention API
+            value_states (`torch.Tensor`):
+                Input value states to be passed to Flash Attention API
+            attention_mask (`torch.Tensor`):
+                The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the
+                position of padding tokens and 1 for the position of non-padding tokens.
+            dropout (`int`, *optional*):
+                Attention dropout
+            softmax_scale (`float`, *optional*):
+                The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim)
+            use_sliding_windows (`bool`, *optional*):
+                Whether to activate sliding window attention.
+        """
+        if not self._flash_attn_uses_top_left_mask:
+            causal = self.is_causal
+        else:
+            # TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in LlamaFlashAttention2 __init__.
+            causal = self.is_causal and query_length != 1
+
+        # Contains at least one padding token in the sequence
+        if attention_mask is not None:
+            batch_size = query_states.shape[0]
+            query_states, key_states, value_states, indices_q, cu_seq_lens, max_seq_lens = self._upad_input(
+                query_states, key_states, value_states, attention_mask, query_length
+            )
+
+            cu_seqlens_q, cu_seqlens_k = cu_seq_lens
+            max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
+
+            if not use_sliding_windows:
+                attn_output_unpad = flash_attn_varlen_func(
+                    query_states,
+                    key_states,
+                    value_states,
+                    cu_seqlens_q=cu_seqlens_q,
+                    cu_seqlens_k=cu_seqlens_k,
+                    max_seqlen_q=max_seqlen_in_batch_q,
+                    max_seqlen_k=max_seqlen_in_batch_k,
+                    dropout_p=dropout,
+                    softmax_scale=softmax_scale,
+                    causal=causal,
+                )
+            else:
+                attn_output_unpad = flash_attn_varlen_func(
+                    query_states,
+                    key_states,
+                    value_states,
+                    cu_seqlens_q=cu_seqlens_q,
+                    cu_seqlens_k=cu_seqlens_k,
+                    max_seqlen_q=max_seqlen_in_batch_q,
+                    max_seqlen_k=max_seqlen_in_batch_k,
+                    dropout_p=dropout,
+                    softmax_scale=softmax_scale,
+                    causal=causal,
+                    window_size=(self.config.sliding_window, self.config.sliding_window),
+                )
+
+            attn_output = pad_input(attn_output_unpad, indices_q, batch_size, query_length)
+        else:
+            if not use_sliding_windows:
+                attn_output = flash_attn_func(
+                    query_states,
+                    key_states,
+                    value_states,
+                    dropout,
+                    softmax_scale=softmax_scale,
+                    causal=causal,
+                )
+            else:
+                attn_output = flash_attn_func(
+                    query_states,
+                    key_states,
+                    value_states,
+                    dropout,
+                    softmax_scale=softmax_scale,
+                    causal=causal,
+                    window_size=(self.config.sliding_window, self.config.sliding_window),
+                )
+
+        return attn_output
+
+    def _upad_input(self, query_layer, key_layer, value_layer, attention_mask, query_length):
+        batch_size, kv_seq_len, num_heads, head_dim = key_layer.shape
+
+        # On the first iteration we need to properly re-create the padding mask
+        # by slicing it on the proper place
+        if kv_seq_len != attention_mask.shape[-1]:
+            attention_mask_num_tokens = attention_mask.shape[-1]
+            attention_mask = attention_mask[:, attention_mask_num_tokens - kv_seq_len :]
+
+        indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask)
+
+        key_layer = index_first_axis(key_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim), indices_k)
+        value_layer = index_first_axis(value_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim), indices_k)
+
+        if query_length == kv_seq_len:
+            query_layer = index_first_axis(
+                query_layer.reshape(batch_size * kv_seq_len, num_heads, head_dim), indices_k
+            )
+            cu_seqlens_q = cu_seqlens_k
+            max_seqlen_in_batch_q = max_seqlen_in_batch_k
+            indices_q = indices_k
+        elif query_length == 1:
+            max_seqlen_in_batch_q = 1
+            cu_seqlens_q = torch.arange(
+                batch_size + 1, dtype=torch.int32, device=query_layer.device
+            )  # There is a memcpy here, that is very bad.
+            indices_q = cu_seqlens_q[:-1]
+            query_layer = query_layer.squeeze(1)
+        else:
+            # The -q_len: slice assumes left padding.
+            attention_mask = attention_mask[:, -query_length:]
+            query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(query_layer, attention_mask)
+
+        return (
+            query_layer,
+            key_layer,
+            value_layer,
+            indices_q,
+            (cu_seqlens_q, cu_seqlens_k),
+            (max_seqlen_in_batch_q, max_seqlen_in_batch_k),
+        )
+
+
+# copied from transformers.models.llama.modeling_llama.LlamaSdpaAttention with Llama->Mistral
+# TODO @Arthur no longer copied from LLama after static cache
+class MistralSdpaAttention(MistralAttention):
+    """
+    Mistral attention module using torch.nn.functional.scaled_dot_product_attention. This module inherits from
+    `MistralAttention` as the weights of the module stays untouched. The only changes are on the forward pass to adapt to
+    SDPA API.
+    """
+
+    # Adapted from MistralAttention.forward
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Cache] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        if output_attentions:
+            # TODO: Improve this warning with e.g. `model.config.attn_implementation = "manual"` once this is implemented.
+            logger.warning_once(
+                "MistralModel is using MistralSdpaAttention, but `torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to the manual attention implementation, "
+                'but specifying the manual implementation will be required from Transformers version v5.0.0 onwards. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
+            )
+            return super().forward(
+                hidden_states=hidden_states,
+                attention_mask=attention_mask,
+                position_ids=position_ids,
+                past_key_value=past_key_value,
+                output_attentions=output_attentions,
+                use_cache=use_cache,
+            )
+
+        bsz, q_len, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states)
+        value_states = self.v_proj(hidden_states)
+
+        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+        value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            kv_seq_len += past_key_value.get_seq_length(self.layer_idx)
+        cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
+
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
+
+        if past_key_value is not None:
+            cache_kwargs = {"sin": sin, "cos": cos}  # Specific to RoPE models
+            key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+        key_states = repeat_kv(key_states, self.num_key_value_groups)
+        value_states = repeat_kv(value_states, self.num_key_value_groups)
+
+        if attention_mask is not None:
+            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
+                raise ValueError(
+                    f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
+                )
+
+        # SDPA with memory-efficient backend is currently (torch==2.1.2) bugged with non-contiguous inputs with custom attn_mask,
+        # Reference: https://github.com/pytorch/pytorch/issues/112577.
+        if query_states.device.type == "cuda" and attention_mask is not None:
+            query_states = query_states.contiguous()
+            key_states = key_states.contiguous()
+            value_states = value_states.contiguous()
+
+        attn_output = torch.nn.functional.scaled_dot_product_attention(
+            query_states,
+            key_states,
+            value_states,
+            attn_mask=attention_mask,
+            dropout_p=self.attention_dropout if self.training else 0.0,
+            # The q_len > 1 is necessary to match with AttentionMaskConverter.to_causal_4d that does not create a causal mask in case q_len == 1.
+            is_causal=self.is_causal and attention_mask is None and q_len > 1,
+        )
+
+        attn_output = attn_output.transpose(1, 2).contiguous()
+        attn_output = attn_output.view(bsz, q_len, self.hidden_size)
+
+        attn_output = self.o_proj(attn_output)
+
+        return attn_output, None, past_key_value
+
+
+MISTRAL_ATTENTION_CLASSES = {
+    "eager": MistralAttention,
+    "flash_attention_2": MistralFlashAttention2,
+    "sdpa": MistralSdpaAttention,
+}
+
+
+class MistralDecoderLayer(nn.Module):
+    def __init__(self, config: MistralConfig, layer_idx: int):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+
+        self.self_attn = MISTRAL_ATTENTION_CLASSES[config._attn_implementation](config, layer_idx)
+
+        self.mlp = MistralMLP(config)
+        self.input_layernorm = MistralRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = MistralRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        output_attentions: Optional[bool] = False,
+        use_cache: Optional[bool] = False,
+        **kwargs,
+    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        if "padding_mask" in kwargs:
+            warnings.warn(
+                "Passing `padding_mask` is deprecated and will be removed in v4.37. Please make sure use `attention_mask` instead.`"
+            )
+        """
+        Args:
+            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
+            attention_mask (`torch.FloatTensor`, *optional*): attention mask of size
+                `(batch, sequence_length)` where padding elements are indicated by 0.
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+            use_cache (`bool`, *optional*):
+                If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
+                (see `past_key_values`).
+            past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
+        """
+
+        residual = hidden_states
+
+        hidden_states = self.input_layernorm(hidden_states)
+
+        # Self Attention
+        hidden_states, self_attn_weights, present_key_value = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_value=past_key_value,
+            output_attentions=output_attentions,
+            use_cache=use_cache,
+        )
+        hidden_states = residual + hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (self_attn_weights,)
+
+        if use_cache:
+            outputs += (present_key_value,)
+
+        return outputs
+
+
+MISTRAL_START_DOCSTRING = r"""
+    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
+    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
+    etc.)
+
+    This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
+    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
+    and behavior.
+
+    Parameters:
+        config ([`MistralConfig`]):
+            Model configuration class with all the parameters of the model. Initializing with a config file does not
+            load the weights associated with the model, only the configuration. Check out the
+            [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+
+@add_start_docstrings(
+    "The bare Mistral Model outputting raw hidden-states without any specific head on top.",
+    MISTRAL_START_DOCSTRING,
+)
+class MistralPreTrainedModel(PreTrainedModel):
+    config_class = MistralConfig
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["MistralDecoderLayer"]
+    _skip_keys_device_placement = "past_key_values"
+    _supports_flash_attn_2 = True
+    _supports_sdpa = True
+    _supports_cache_class = True
+
+    def _init_weights(self, module):
+        std = self.config.initializer_range
+        if isinstance(module, nn.Linear):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+
+
+MISTRAL_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
+            Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
+            it.
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input-ids)
+        attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+
+            [What are attention masks?](../glossary#attention-mask)
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            If `past_key_values` is used, optionally only the last `decoder_input_ids` have to be input (see
+            `past_key_values`).
+
+            If you want to change padding behavior, you should read [`modeling_opt._prepare_decoder_attention_mask`]
+            and modify to your needs. See diagram 1 in [the paper](https://arxiv.org/abs/1910.13461) for more
+            information on the default strategy.
+
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+        position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
+            config.n_positions - 1]`.
+
+            [What are position IDs?](../glossary#position-ids)
+        past_key_values (`Cache` or `tuple(tuple(torch.FloatTensor))`, *optional*):
+            Pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
+            blocks) that can be used to speed up sequential decoding. This typically consists in the `past_key_values`
+            returned by the model at a previous stage of decoding, when `use_cache=True` or `config.use_cache=True`.
+
+            Two formats are allowed:
+            - a [`~cache_utils.Cache`] instance;
+            - Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of
+            shape `(batch_size, num_heads, sequence_length, embed_size_per_head)`). This is also known as the legacy
+            cache format.
+
+            The model will output the same cache format that is fed as input. If no `past_key_values` are passed, the
+            legacy cache format will be returned.
+
+            If `past_key_values` are used, the user can optionally input only the last `input_ids` (those that don't
+            have their past key value states given to this model) of shape `(batch_size, 1)` instead of all `input_ids`
+            of shape `(batch_size, sequence_length)`.
+        inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
+            is useful if you want more control over how to convert `input_ids` indices into associated vectors than the
+            model's internal embedding lookup matrix.
+        use_cache (`bool`, *optional*):
+            If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see
+            `past_key_values`).
+        output_attentions (`bool`, *optional*):
+            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
+            tensors for more detail.
+        output_hidden_states (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+            more detail.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+"""
+
+
+@add_start_docstrings(
+    "The bare Mistral Model outputting raw hidden-states without any specific head on top.",
+    MISTRAL_START_DOCSTRING,
+)
+class MistralModel(MistralPreTrainedModel):
+    """
+    Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`MistralDecoderLayer`]
+
+    Args:
+        config: MistralConfig
+    """
+
+    def __init__(self, config: MistralConfig):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+        self.is_causal = config.is_causal
+
+        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
+        self.layers = nn.ModuleList(
+            [MistralDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
+        )
+        self._attn_implementation = config._attn_implementation
+        self.norm = MistralRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+        self.gradient_checkpointing = False
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.embed_tokens = value
+
+    @add_start_docstrings_to_model_forward(MISTRAL_INPUTS_DOCSTRING)
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        # retrieve input_ids and inputs_embeds
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time")
+        elif input_ids is not None:
+            batch_size, seq_length = input_ids.shape
+        elif inputs_embeds is not None:
+            batch_size, seq_length, _ = inputs_embeds.shape
+        else:
+            raise ValueError("You have to specify either decoder_input_ids or decoder_inputs_embeds")
+
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                logger.warning_once(
+                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                )
+                use_cache = False
+
+        past_key_values_length = 0
+
+        if use_cache:
+            use_legacy_cache = not isinstance(past_key_values, Cache)
+            if use_legacy_cache:
+                past_key_values = DynamicCache.from_legacy_cache(past_key_values)
+            past_key_values_length = past_key_values.get_seq_length()
+
+        if position_ids is None:
+            device = input_ids.device if input_ids is not None else inputs_embeds.device
+            position_ids = torch.arange(
+                past_key_values_length, seq_length + past_key_values_length, dtype=torch.long, device=device
+            )
+            position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
+        else:
+            # Use position_ids' own last dim to avoid shape mismatch during generation
+            # (seq_length reflects original input length, but position_ids may be trimmed)
+            position_ids = position_ids.view(-1, position_ids.shape[-1]).long()
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+
+        if attention_mask is not None and self._attn_implementation == "flash_attention_2" and use_cache:
+            is_padding_right = attention_mask[:, -1].sum().item() != batch_size
+            if is_padding_right:
+                raise ValueError(
+                    "You are attempting to perform batched generation with padding_side='right'"
+                    " this may lead to unexpected behaviour for Flash Attention version of Mistral. Make sure to "
+                    " call `tokenizer.padding_side  = 'left'` before tokenizing the input. "
+                )
+                
+
+        if self._attn_implementation == "flash_attention_2":
+            # 2d mask is passed through the layers
+            attention_mask = attention_mask if (attention_mask is not None and 0 in attention_mask) else None
+        elif self._attn_implementation == "sdpa" and not output_attentions:
+            # output_attentions=True can not be supported when using SDPA, and we fall back on
+            # the manual implementation that requires a 4D causal mask in all cases.
+            if self.is_causal:
+                attention_mask = _prepare_4d_causal_attention_mask_for_sdpa(
+                    attention_mask,
+                    (batch_size, seq_length),
+                    inputs_embeds,
+                    past_key_values_length,
+                ) 
+            else:
+                attention_mask = _prepare_4d_attention_mask_for_sdpa(
+                    attention_mask,
+                    dtype=inputs_embeds.dtype,
+                )
+        else:
+            # 4d mask is passed through the layers
+            if self.is_causal:
+                attention_mask = _prepare_4d_causal_attention_mask(
+                    attention_mask,
+                    (batch_size, seq_length),
+                    inputs_embeds,
+                    past_key_values_length,
+                    sliding_window=self.config.sliding_window,
+                )
+            else:
+                attention_mask = _prepare_4d_attention_mask(
+                    attention_mask,
+                    dtype=inputs_embeds.dtype,
+                )
+        
+        hidden_states = inputs_embeds
+
+        # decoder layers
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+        next_decoder_cache = None
+
+        for decoder_layer in self.layers:
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            if self.gradient_checkpointing and self.training:
+                layer_outputs = self._gradient_checkpointing_func(
+                    decoder_layer.__call__,
+                    hidden_states,
+                    attention_mask,
+                    position_ids,
+                    past_key_values,
+                    output_attentions,
+                    use_cache,
+                )
+            else:
+                layer_outputs = decoder_layer(
+                    hidden_states,
+                    attention_mask=attention_mask,
+                    position_ids=position_ids,
+                    past_key_value=past_key_values,
+                    output_attentions=output_attentions,
+                    use_cache=use_cache,
+                )
+
+            hidden_states = layer_outputs[0]
+
+            if use_cache:
+                next_decoder_cache = layer_outputs[2 if output_attentions else 1]
+
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+        hidden_states = self.norm(hidden_states)
+
+        # add hidden states from the last decoder layer
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        next_cache = None
+        if use_cache:
+            next_cache = next_decoder_cache.to_legacy_cache() if use_legacy_cache else next_decoder_cache
+
+        if not return_dict:
+            return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None)
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=next_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+        )
+
+
+class MistralForCausalLM(MistralPreTrainedModel):
+    _tied_weights_keys = ["lm_head.weight"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = MistralModel(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.model.embed_tokens = value
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def set_decoder(self, decoder):
+        self.model = decoder
+
+    def get_decoder(self):
+        return self.model
+
+    @add_start_docstrings_to_model_forward(MISTRAL_INPUTS_DOCSTRING)
+    @replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, CausalLMOutputWithPast]:
+        r"""
+        Args:
+            labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+                config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+                (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        Returns:
+
+        Example:
+
+        ```python
+        >>> from transformers import AutoTokenizer, MistralForCausalLM
+
+        >>> model = MistralForCausalLM.from_pretrained("mistralai/Mistral-7B-v0.1")
+        >>> tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1")
+
+        >>> prompt = "Hey, are you conscious? Can you talk to me?"
+        >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+        >>> # Generate
+        >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+        "Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
+        ```"""
+
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        hidden_states = outputs[0]
+        logits = self.lm_head(hidden_states)
+        logits = logits.float()
+
+        loss = None
+        if labels is not None:
+            # Shift so that tokens < n predict n
+            shift_logits = logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
+            loss_fct = CrossEntropyLoss()
+            shift_logits = shift_logits.view(-1, self.config.vocab_size)
+            shift_labels = shift_labels.view(-1)
+            # Enable model parallelism
+            shift_labels = shift_labels.to(shift_logits.device)
+            loss = loss_fct(shift_logits, shift_labels)
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+    def prepare_inputs_for_generation(
+        self, input_ids, past_key_values=None, attention_mask=None, inputs_embeds=None, **kwargs
+    ):
+        # Omit tokens covered by past_key_values
+        if past_key_values is not None:
+            if isinstance(past_key_values, Cache):
+                cache_length = past_key_values.get_seq_length()
+                past_length = past_key_values.get_seq_length()
+                max_cache_length = past_key_values.get_max_cache_shape()
+            else:
+                cache_length = past_length = past_key_values[0][0].shape[2]
+                max_cache_length = None
+
+            # Keep only the unprocessed tokens:
+            # 1 - If the length of the attention_mask exceeds the length of input_ids, then we are in a setting where
+            # some of the inputs are exclusively passed as part of the cache (e.g. when passing input_embeds as
+            # input)
+            if attention_mask is not None and attention_mask.shape[1] > input_ids.shape[1]:
+                input_ids = input_ids[:, -(attention_mask.shape[1] - past_length) :]
+            # 2 - If the past_length is smaller than input_ids', then input_ids holds all input tokens. We can discard
+            # input_ids based on the past_length.
+            elif past_length < input_ids.shape[1]:
+                input_ids = input_ids[:, past_length:]
+            # 3 - Otherwise (past_length >= input_ids.shape[1]), let's assume input_ids only has unprocessed tokens.
+
+            # If we are about to go beyond the maximum cache length, we need to crop the input attention mask.
+            if (
+                max_cache_length is not None
+                and attention_mask is not None
+                and cache_length + input_ids.shape[1] > max_cache_length
+            ):
+                attention_mask = attention_mask[:, -max_cache_length:]
+
+        position_ids = kwargs.get("position_ids", None)
+        if attention_mask is not None and position_ids is None:
+            # create position_ids on the fly for batch generation
+            position_ids = attention_mask.long().cumsum(-1) - 1
+            position_ids.masked_fill_(attention_mask == 0, 1)
+            if past_key_values:
+                position_ids = position_ids[:, -input_ids.shape[1] :]
+
+        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and past_key_values is None:
+            model_inputs = {"inputs_embeds": inputs_embeds}
+        else:
+            model_inputs = {"input_ids": input_ids}
+
+        model_inputs.update(
+            {
+                "position_ids": position_ids,
+                "past_key_values": past_key_values,
+                "use_cache": kwargs.get("use_cache"),
+                "attention_mask": attention_mask,
+            }
+        )
+        return model_inputs
+
+    @staticmethod
+    def _reorder_cache(past_key_values, beam_idx):
+        reordered_past = ()
+        for layer_past in past_key_values:
+            reordered_past += (
+                tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past),
+            )
+        return reordered_past
+
+
+
+
+@dataclass
+class MoEMaskedLMOutput(ModelOutput):
+    """
+    Base class for causal language model (or autoregressive) outputs as well as Mixture of Expert's router hidden
+    states terms, to train a MoE model.
+
+    Args:
+        loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
+            Language modeling loss (for next-token prediction).
+        logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`):
+            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
+        past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
+            `(batch_size, num_heads, sequence_length, embed_size_per_head)`)
+
+            Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see
+            `past_key_values` input) to speed up sequential decoding.
+        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, +
+            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
+        attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+        z_loss (`torch.FloatTensor`, *optional*, returned when `labels` is provided):
+            z_loss for the sparse modules.
+        aux_loss (`torch.FloatTensor`, *optional*, returned when `labels` is provided):
+            aux_loss for the sparse modules.
+        router_logits (`tuple(torch.FloatTensor)`, *optional*, returned when `output_router_logits=True` is passed or when `config.add_router_probs=True`):
+            Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, sequence_length, num_experts)`.
+
+            Router logits of the encoder model, useful to compute the auxiliary loss and the z_loss for the sparse
+            modules.
+    """
+
+    loss: Optional[torch.FloatTensor] = None
+    logits: torch.FloatTensor = None
+    hidden_states: Optional[Tuple[torch.FloatTensor, ...]] = None
+    attentions: Optional[Tuple[torch.FloatTensor, ...]] = None
+    z_loss: torch.FloatTensor = None
+    aux_loss: torch.FloatTensor = None
+    router_logits: Optional[Tuple[torch.FloatTensor]] = None
+
+
+
+class MistralPredictionHeadTransform(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        if isinstance(config.hidden_act, str):
+            self.transform_act_fn = ACT2FN[config.hidden_act]
+        else:
+            self.transform_act_fn = config.hidden_act
+        self.norm = MistralRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.transform_act_fn(hidden_states)
+        hidden_states = self.norm(hidden_states)
+        return hidden_states
+
+
+class MistralLMPredictionHead(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.transform = MistralPredictionHeadTransform(config)
+
+        # The output weights are the same as the input embeddings, but there is
+        # an output-only bias for each token.
+        self.decoder = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        self.bias = nn.Parameter(torch.zeros(config.vocab_size))
+
+        # Need a link between the two variables so that the bias is correctly resized with `resize_token_embeddings`
+        self.decoder.bias = self.bias
+
+    def _tie_weights(self):
+        self.decoder.bias = self.bias
+
+    def forward(self, hidden_states):
+        hidden_states = self.transform(hidden_states)
+        hidden_states = self.decoder(hidden_states)
+        return hidden_states
+
+
+class MistralOnlyMLMHead(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.predictions = MistralLMPredictionHead(config)
+
+    def forward(self, sequence_output: torch.Tensor) -> torch.Tensor:
+        prediction_scores = self.predictions(sequence_output)
+        return prediction_scores
+
+
+
+class MistralForMaskedLM(MistralPreTrainedModel):
+    _tied_weights_keys = ["predictions.decoder.bias", "cls.predictions.decoder.weight"]
+
+    def __init__(self, config):
+        super().__init__(config)
+
+        if config.is_decoder:
+            logger.warning(
+                "If you want to use `MistralForMaskedLM` make sure `config.is_decoder=False` for "
+                "bi-directional self-attention."
+            )
+
+        self.model = MistralModel(config)
+        self.cls = MistralOnlyMLMHead(config)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_output_embeddings(self):
+        return self.cls.predictions.decoder
+
+    def set_output_embeddings(self, new_embeddings):
+        self.cls.predictions.decoder = new_embeddings
+        self.cls.predictions.bias = new_embeddings.bias
+
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        labels: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple[torch.Tensor], MoEMaskedLMOutput]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should be in `[-100, 0, ...,
+            config.vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are ignored (masked), the
+            loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`
+        """
+
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.model(
+            input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            inputs_embeds=inputs_embeds,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        
+
+        sequence_output = outputs[0]
+        logits = self.cls(sequence_output)
+
+        loss = None
+        if labels is not None:
+            loss_fct = CrossEntropyLoss()  # -100 index = padding token
+            loss = loss_fct(logits.view(-1, self.config.vocab_size), labels.view(-1))
+            
+
+        if not return_dict:
+            output = (logits,) + outputs[2:]
+            return ((loss,) + output) if loss is not None else output
+
+        return MoEMaskedLMOutput(
+            loss=loss,
+            logits=logits,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+
+
+
+@add_start_docstrings(
+    """
+    The Mistral Model transformer with a sequence classification head on top (linear layer).
+
+    [`MistralForSequenceClassification`] uses the last token in order to do the classification, as other causal models
+    (e.g. GPT-2) do.
+
+    Since it does classification on the last token, it requires to know the position of the last token. If a
+    `pad_token_id` is defined in the configuration, it finds the last token that is not a padding token in each row. If
+    no `pad_token_id` is defined, it simply takes the last value in each row of the batch. Since it cannot guess the
+    padding tokens when `inputs_embeds` are passed instead of `input_ids`, it does the same (take the last value in
+    each row of the batch).
+    """,
+    MISTRAL_START_DOCSTRING,
+)
+# Copied from transformers.models.llama.modeling_llama.LlamaForSequenceClassification with Llama->Mistral, LLAMA->MISTRAL
+class MistralForSequenceClassification(MistralPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+        self.model = MistralModel(config)
+        self.score = nn.Linear(config.hidden_size, self.num_labels, bias=False)
+        self.dropout = nn.Dropout(config.classifier_dropout)
+        self.is_causal = config.is_causal
+
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.model.embed_tokens = value
+
+    @add_start_docstrings_to_model_forward(MISTRAL_INPUTS_DOCSTRING)
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, SequenceClassifierOutputWithPast]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
+            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
+            `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        transformer_outputs = self.model(
+            input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        hidden_states = transformer_outputs[0]
+        hidden_states = self.dropout(hidden_states)
+        logits = self.score(hidden_states)
+
+        if input_ids is not None:
+            batch_size = input_ids.shape[0]
+        else:
+            batch_size = inputs_embeds.shape[0]
+
+        if self.is_causal:
+            if self.config.pad_token_id is None and batch_size != 1:
+                raise ValueError("Cannot handle batch sizes > 1 if no padding token is defined.")
+            if self.config.pad_token_id is None:
+                sequence_lengths = -1
+            else:
+                if input_ids is not None:
+                    # if no pad token found, use modulo instead of reverse indexing for ONNX compatibility
+                    sequence_lengths = torch.eq(input_ids, self.config.pad_token_id).int().argmax(-1) - 1
+                    sequence_lengths = sequence_lengths % input_ids.shape[-1]
+                    sequence_lengths = sequence_lengths.to(logits.device)
+                else:
+                    sequence_lengths = -1
+
+            pooled_logits = logits[torch.arange(batch_size, device=logits.device), sequence_lengths]
+        else:
+            pooled_logits = logits[:, 0]
+
+        loss = None
+        if labels is not None:
+            labels = labels.to(logits.device)
+            if self.config.problem_type is None:
+                if self.num_labels == 1:
+                    self.config.problem_type = "regression"
+                elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
+                    self.config.problem_type = "single_label_classification"
+                else:
+                    self.config.problem_type = "multi_label_classification"
+
+            if self.config.problem_type == "regression":
+                loss_fct = MSELoss()
+                if self.num_labels == 1:
+                    loss = loss_fct(pooled_logits.squeeze(), labels.squeeze())
+                else:
+                    loss = loss_fct(pooled_logits, labels)
+            elif self.config.problem_type == "single_label_classification":
+                loss_fct = CrossEntropyLoss()
+                loss = loss_fct(pooled_logits.view(-1, self.num_labels), labels.view(-1))
+            elif self.config.problem_type == "multi_label_classification":
+                loss_fct = BCEWithLogitsLoss()
+                loss = loss_fct(pooled_logits, labels)
+        if not return_dict:
+            output = (pooled_logits,) + transformer_outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return SequenceClassifierOutputWithPast(
+            loss=loss,
+            logits=pooled_logits,
+            past_key_values=transformer_outputs.past_key_values,
+            hidden_states=transformer_outputs.hidden_states,
+            attentions=transformer_outputs.attentions,
+        )
+
+# import torch
+# from safetensors import safe_open
+# from safetensors.torch import save_file
+
+# tensors = {}
+# with safe_open("/root/MOE_DNA/trained_model/mistral_mlm_alldata_len5k_ep3/model.safetensors", framework="pt", device="cpu") as f:
+#     # print(f.metadata())
+    
+#     for key in f.keys():
+#         tensors[key] = f.get_tensor(key)
+
+# new_model = {}    
+# for key in tensors.keys():
+#     k = key.replace("Mistral", "model")
+#     new_model[k] = tensors[key]
+

pyproject.toml

@@ -1,28 +0,0 @@
-[build-system]
-requires = ["setuptools>=69"]
-build-backend = "setuptools.build_meta"
-
-[project]
-name = "genomeocean-sub-classifier"
-version = "0.1.0"
-description = "GenomeOcean NCLDV versus Mirus classifier for genomic FASTA files"
-requires-python = ">=3.11"
-dependencies = [
-    "torch>=2.8,<2.9",
-    "transformers==4.51.3",
-    "huggingface-hub>=0.36,<1",
-    "safetensors>=0.5,<1",
-    "numpy>=2.2,<3",
-    "pandas>=2.2,<3",
-]
-
-[project.scripts]
-genomeocean-sub = "genomeocean_sub.cli:main"
-
-[project.optional-dependencies]
-dev = [
-    "pytest>=8,<9",
-]
-
-[tool.setuptools.packages.find]
-where = ["src"]

src/genomeocean_sub/__init__.py

@@ -1,3 +0,0 @@
-"""User-facing inference package for the GenomeOcean sub classifier."""
-
-__version__ = "0.1.0"

src/genomeocean_sub/aggregation.py

@@ -1,113 +0,0 @@
-"""Aggregate Sub predictions from chunks to contigs and files."""
-
-from __future__ import annotations
-
-from collections import Counter
-from typing import Iterable, Mapping
-
-import pandas as pd
-
-
-SUB_LABELS = {0: "NCLDV", 1: "Mirus"}
-PROBABILITY_COLUMNS = {0: "prob_ncldv", 1: "prob_mirus"}
-COUNT_COLUMNS = {0: "n_chunks_ncldv", 1: "n_chunks_mirus"}
-
-
-def _winner(values: Iterable[int], labels: Mapping[int, str]) -> tuple[int, int]:
-    counts = Counter(int(value) for value in values)
-    invalid = set(counts) - set(labels)
-    if invalid:
-        raise ValueError(f"Unexpected predicted labels: {sorted(invalid)}")
-    winner = min(labels, key=lambda label: (-counts[label], label))
-    return winner, counts[winner]
-
-
-def aggregate_chunks_to_contigs(chunk_results: pd.DataFrame) -> pd.DataFrame:
-    output_columns = [
-        "record_id",
-        "source_file",
-        "contig_id",
-        "original_length",
-        "clean_length",
-        "n_chunks",
-        "ignored_tail_bp",
-        *COUNT_COLUMNS.values(),
-        *PROBABILITY_COLUMNS.values(),
-        "ensemble_size",
-        "mean_chunk_ensemble_agreement",
-        "mean_chunk_confidence_std",
-        "predicted_label",
-        "predicted_name",
-        "predicted_votes",
-        "confidence",
-    ]
-    if chunk_results.empty:
-        return pd.DataFrame(columns=output_columns)
-
-    group_columns = [
-        "record_id",
-        "source_file",
-        "contig_id",
-        "original_length",
-        "clean_length",
-        "n_chunks",
-        "ignored_tail_bp",
-    ]
-    rows: list[dict] = []
-    for group_key, group in chunk_results.groupby(group_columns, sort=False, dropna=False):
-        row = dict(zip(group_columns, group_key))
-        winner, votes = _winner(group["predicted_label"], SUB_LABELS)
-        counts = Counter(group["predicted_label"].astype(int))
-        for label, column in COUNT_COLUMNS.items():
-            row[column] = int(counts[label])
-        for _, column in PROBABILITY_COLUMNS.items():
-            row[column] = float(group[column].mean())
-        row["ensemble_size"] = int(
-            group["ensemble_size"].max() if "ensemble_size" in group else 1
-        )
-        row["mean_chunk_ensemble_agreement"] = float(
-            group["ensemble_agreement"].mean()
-            if "ensemble_agreement" in group
-            else 1.0
-        )
-        row["mean_chunk_confidence_std"] = float(
-            group["confidence_std"].mean() if "confidence_std" in group else 0.0
-        )
-        row["predicted_label"] = winner
-        row["predicted_name"] = SUB_LABELS[winner]
-        row["predicted_votes"] = votes
-        row["confidence"] = row[PROBABILITY_COLUMNS[winner]]
-        rows.append(row)
-    return pd.DataFrame(rows, columns=output_columns)
-
-
-def aggregate_contigs_to_files(contig_results: pd.DataFrame) -> pd.DataFrame:
-    count_columns = {0: "n_contigs_ncldv", 1: "n_contigs_mirus"}
-    output_columns = [
-        "source_file",
-        "n_contigs",
-        *count_columns.values(),
-        "predicted_label",
-        "predicted_name",
-        "predicted_votes",
-        "confidence",
-    ]
-    if contig_results.empty:
-        return pd.DataFrame(columns=output_columns)
-
-    rows: list[dict] = []
-    for source_file, group in contig_results.groupby("source_file", sort=False):
-        winner, votes = _winner(group["predicted_label"], SUB_LABELS)
-        counts = Counter(group["predicted_label"].astype(int))
-        row = {
-            "source_file": source_file,
-            "n_contigs": len(group),
-            "predicted_label": winner,
-            "predicted_name": SUB_LABELS[winner],
-            "predicted_votes": votes,
-            "confidence": float(group.loc[group["predicted_label"] == winner, "confidence"].mean()),
-        }
-        for label, column in count_columns.items():
-            row[column] = int(counts[label])
-        rows.append(row)
-    return pd.DataFrame(rows, columns=output_columns)

src/genomeocean_sub/cli.py

@@ -1,85 +0,0 @@
-"""Command-line interface for the Sub classifier."""
-
-from __future__ import annotations
-
-import argparse
-import os
-
-from .predict import SubPredictor
-
-
-def build_parser() -> argparse.ArgumentParser:
-    parser = argparse.ArgumentParser(
-        prog="genomeocean-sub",
-        description="Classify FASTA contigs as NCLDV or Mirus.",
-    )
-    subparsers = parser.add_subparsers(dest="command", required=True)
-    predict = subparsers.add_parser("predict", help="classify a FASTA file or directory")
-    predict.add_argument("--input", required=True)
-    predict.add_argument("--output-dir", required=True)
-    predict.add_argument(
-        "--model-id",
-        action="append",
-        default=None,
-        help=(
-            "Hugging Face model ID or local model path. Repeat this option for "
-            "independent fold paths. A single value may also be set with "
-            "GENOMEOCEAN_SUB_MODEL."
-        ),
-    )
-    predict.add_argument(
-        "--subfolder",
-        action="append",
-        default=None,
-        help=(
-            "Fold subfolder inside one shared --model-id, for example fold1. "
-            "Repeat for all ensemble members."
-        ),
-    )
-    predict.add_argument("--revision", default=None)
-    predict.add_argument("--device", default="auto")
-    predict.add_argument("--batch-size", type=int, default=8)
-    predict.add_argument("--chunk-size", type=int, default=5000)
-    predict.add_argument("--stride", type=int, default=5000)
-    predict.add_argument("--max-length", type=int, default=1250)
-    predict.add_argument("--cache-dir", default=None)
-    predict.add_argument("--local-files-only", action="store_true")
-    return parser
-
-
-def main() -> None:
-    args = build_parser().parse_args()
-    if args.command == "predict":
-        model_ids = args.model_id
-        if not model_ids:
-            environment_model = os.environ.get("GENOMEOCEAN_SUB_MODEL")
-            model_ids = [environment_model] if environment_model else None
-        if not model_ids:
-            raise SystemExit(
-                "--model-id is required until a public default model has been published"
-            )
-        predictor = SubPredictor(
-            model_ids,
-            revision=args.revision,
-            subfolders=args.subfolder,
-            device=args.device,
-            batch_size=args.batch_size,
-            max_length=args.max_length,
-            cache_dir=args.cache_dir,
-            local_files_only=args.local_files_only,
-        )
-        bundle = predictor.predict_fasta(
-            args.input,
-            args.output_dir,
-            chunk_size=args.chunk_size,
-            stride=args.stride,
-        )
-        print(
-            f"[OK] predicted {len(bundle.contig_results)} contigs; "
-            f"ensemble models {len(predictor.model_specs)}; "
-            f"skipped {len(bundle.skipped)}; outputs: {args.output_dir}"
-        )
-
-
-if __name__ == "__main__":
-    main()

src/genomeocean_sub/fasta.py

@@ -1,125 +0,0 @@
-"""Small FASTA reader/writer for the independent Sub classifier."""
-
-from __future__ import annotations
-
-from dataclasses import dataclass
-import gzip
-from pathlib import Path
-from typing import Iterable, Iterator, TextIO
-
-
-FASTA_SUFFIXES = (".fa", ".fasta", ".fna", ".fa.gz", ".fasta.gz", ".fna.gz")
-
-
-@dataclass(frozen=True)
-class FastaRecord:
-    source_file: str
-    contig_id: str
-    description: str
-    sequence: str
-
-    @property
-    def record_id(self) -> str:
-        return f"{self.source_file}::{self.contig_id}"
-
-
-def is_fasta_path(path: Path) -> bool:
-    return path.is_file() and path.name.lower().endswith(FASTA_SUFFIXES)
-
-
-def discover_fasta_files(input_path: str | Path) -> list[Path]:
-    path = Path(input_path).expanduser().resolve()
-    if not path.exists():
-        raise FileNotFoundError(f"Input does not exist: {path}")
-    if path.is_file():
-        if not is_fasta_path(path):
-            raise ValueError(
-                f"Unsupported FASTA extension: {path.name}. "
-                f"Expected one of: {', '.join(FASTA_SUFFIXES)}"
-            )
-        return [path]
-    files = sorted(candidate for candidate in path.rglob("*") if is_fasta_path(candidate))
-    if not files:
-        raise ValueError(f"No FASTA files found under: {path}")
-    return files
-
-
-def _open_text(path: Path) -> TextIO:
-    if path.name.lower().endswith(".gz"):
-        return gzip.open(path, "rt")
-    return path.open("rt")
-
-
-def read_fasta(path: str | Path, *, source_name: str | None = None) -> Iterator[FastaRecord]:
-    fasta_path = Path(path)
-    source_file = source_name or fasta_path.name
-    seen_ids: set[str] = set()
-    description: str | None = None
-    sequence_parts: list[str] = []
-
-    def build_record() -> FastaRecord:
-        assert description is not None
-        contig_id = description.split()[0]
-        if contig_id in seen_ids:
-            raise ValueError(f"Duplicate FASTA ID '{contig_id}' in {fasta_path}")
-        seen_ids.add(contig_id)
-        return FastaRecord(
-            source_file=source_file,
-            contig_id=contig_id,
-            description=description,
-            sequence="".join(sequence_parts),
-        )
-
-    with _open_text(fasta_path) as handle:
-        for line_number, raw_line in enumerate(handle, start=1):
-            line = raw_line.strip()
-            if not line:
-                continue
-            if line.startswith(">"):
-                if description is not None:
-                    yield build_record()
-                description = line[1:].strip()
-                if not description:
-                    raise ValueError(f"Empty FASTA header at {fasta_path}:{line_number}")
-                sequence_parts = []
-            else:
-                if description is None:
-                    raise ValueError(
-                        f"Sequence appears before the first FASTA header at "
-                        f"{fasta_path}:{line_number}"
-                    )
-                sequence_parts.append(line)
-    if description is not None:
-        yield build_record()
-
-
-def read_input_records(input_path: str | Path) -> list[FastaRecord]:
-    input_root = Path(input_path).expanduser().resolve()
-    files = discover_fasta_files(input_root)
-    base = input_root if input_root.is_dir() else input_root.parent
-    records: list[FastaRecord] = []
-    for path in files:
-        records.extend(
-            read_fasta(path, source_name=path.relative_to(base).as_posix())
-        )
-    if not records:
-        raise ValueError(f"No FASTA records found in: {input_root}")
-    return records
-
-
-def write_fasta(
-    records: Iterable[FastaRecord],
-    output_path: str | Path,
-    *,
-    use_record_id: bool = False,
-    line_width: int = 80,
-) -> Path:
-    path = Path(output_path)
-    path.parent.mkdir(parents=True, exist_ok=True)
-    with path.open("w") as handle:
-        for record in records:
-            header = record.record_id if use_record_id else record.description
-            handle.write(f">{header}\n")
-            for start in range(0, len(record.sequence), line_width):
-                handle.write(record.sequence[start : start + line_width] + "\n")
-    return path

src/genomeocean_sub/predict.py

@@ -1,416 +0,0 @@
-"""Hugging Face model loading and end-to-end NCLDV/Mirus classification."""
-
-from __future__ import annotations
-
-from dataclasses import asdict, dataclass
-import gc
-import json
-from pathlib import Path
-from typing import Sequence
-
-import pandas as pd
-import torch
-from transformers import AutoModelForSequenceClassification, AutoTokenizer
-
-from .aggregation import (
-    PROBABILITY_COLUMNS,
-    SUB_LABELS,
-    aggregate_chunks_to_contigs,
-    aggregate_contigs_to_files,
-)
-from .fasta import FastaRecord, read_input_records
-from .preprocessing import (
-    DEFAULT_CHUNK_SIZE,
-    DEFAULT_STRIDE,
-    SequenceChunk,
-    SkippedRecord,
-    preprocess_records,
-)
-
-ENSEMBLE_METHOD = "mean_probabilities"
-SUB_CHUNK_RESULT_COLUMNS = [
-    "record_id",
-    "source_file",
-    "contig_id",
-    "original_length",
-    "clean_length",
-    "n_chunks",
-    "ignored_tail_bp",
-    "chunk_index",
-    "chunk_start",
-    "chunk_end",
-    "predicted_label",
-    "predicted_name",
-    "confidence",
-    "ensemble_size",
-    "ensemble_votes",
-    "ensemble_agreement",
-    "confidence_std",
-    *PROBABILITY_COLUMNS.values(),
-]
-SKIPPED_RESULT_COLUMNS = [
-    "record_id",
-    "source_file",
-    "contig_id",
-    "original_length",
-    "clean_length",
-    "reason",
-]
-
-
-@dataclass(frozen=True)
-class ModelSpec:
-    """One Sub ensemble member stored locally or on Hugging Face."""
-
-    model_id: str
-    revision: str | None = None
-    subfolder: str | None = None
-    name: str | None = None
-
-    def to_dict(self) -> dict:
-        return asdict(self)
-
-
-@dataclass
-class SubPredictionBundle:
-    records: list[FastaRecord]
-    chunks: list[SequenceChunk]
-    skipped: list[SkippedRecord]
-    chunk_results: pd.DataFrame
-    contig_results: pd.DataFrame
-    file_results: pd.DataFrame
-
-
-def build_model_specs(
-    model_ids: str | Sequence[str],
-    *,
-    revision: str | None = None,
-    subfolders: Sequence[str] | None = None,
-) -> list[ModelSpec]:
-    ids = [model_ids] if isinstance(model_ids, str) else list(model_ids)
-    ids = [str(value).strip() for value in ids if str(value).strip()]
-    if not ids:
-        raise ValueError("At least one model_id is required")
-
-    folders = [str(value).strip() for value in (subfolders or []) if str(value).strip()]
-    if folders:
-        if len(ids) != 1:
-            raise ValueError(
-                "subfolders can only be used with one shared model_id; "
-                "repeat model_id instead when every fold has a different path"
-            )
-        return [
-            ModelSpec(
-                model_id=ids[0],
-                revision=revision,
-                subfolder=folder,
-                name=folder,
-            )
-            for folder in folders
-        ]
-
-    return [
-        ModelSpec(
-            model_id=model_id,
-            revision=revision,
-            name=f"model{index}",
-        )
-        for index, model_id in enumerate(ids, start=1)
-    ]
-
-
-def resolve_device(requested: str) -> torch.device:
-    if requested == "auto":
-        if torch.cuda.is_available():
-            return torch.device("cuda")
-        if hasattr(torch.backends, "mps") and torch.backends.mps.is_available():
-            return torch.device("mps")
-        return torch.device("cpu")
-    device = torch.device(requested)
-    if device.type == "cuda" and not torch.cuda.is_available():
-        raise RuntimeError("CUDA was requested, but torch.cuda.is_available() is False")
-    return device
-
-
-class ProbabilityEnsembleAccumulator:
-    """Accumulate fold outputs and calculate a soft-voting prediction."""
-
-    def __init__(self, n_items: int, n_labels: int):
-        self.n_items = n_items
-        self.n_labels = n_labels
-        self.n_models = 0
-        self._sum = torch.zeros((n_items, n_labels), dtype=torch.float64)
-        self._sum_squares = torch.zeros((n_items, n_labels), dtype=torch.float64)
-        self._votes = torch.zeros((n_items, n_labels), dtype=torch.int64)
-
-    def add(self, probabilities: torch.Tensor) -> None:
-        values = probabilities.detach().to(device="cpu", dtype=torch.float64)
-        expected = (self.n_items, self.n_labels)
-        if tuple(values.shape) != expected:
-            raise ValueError(
-                f"Model probabilities have shape {tuple(values.shape)}; expected {expected}"
-            )
-        if not torch.isfinite(values).all():
-            raise ValueError("Model probabilities contain NaN or infinite values")
-
-        self._sum += values
-        self._sum_squares += values.square()
-        predicted = values.argmax(dim=1)
-        self._votes.scatter_add_(
-            1,
-            predicted.unsqueeze(1),
-            torch.ones((self.n_items, 1), dtype=torch.int64),
-        )
-        self.n_models += 1
-
-    def finalize(
-        self,
-    ) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]:
-        if self.n_models == 0:
-            raise ValueError("Cannot finalize an ensemble with no model predictions")
-
-        mean = self._sum / self.n_models
-        variance = (self._sum_squares / self.n_models - mean.square()).clamp_min(0)
-        standard_deviation = variance.sqrt()
-        predicted = mean.argmax(dim=1)
-        selected_votes = self._votes.gather(1, predicted.unsqueeze(1)).squeeze(1)
-        agreement = selected_votes.to(torch.float64) / self.n_models
-        confidence_std = standard_deviation.gather(
-            1, predicted.unsqueeze(1)
-        ).squeeze(1)
-        return mean, predicted, selected_votes, agreement, confidence_std
-
-
-class SubPredictor:
-    """Predict NCLDV/Mirus with one model or a sequential fold ensemble."""
-
-    def __init__(
-        self,
-        model_id: str | Sequence[str],
-        *,
-        revision: str | None = None,
-        subfolders: Sequence[str] | None = None,
-        device: str = "auto",
-        batch_size: int = 8,
-        max_length: int = 1250,
-        cache_dir: str | None = None,
-        local_files_only: bool = False,
-    ):
-        if batch_size <= 0:
-            raise ValueError("batch_size must be greater than zero")
-        self.model_specs = build_model_specs(
-            model_id,
-            revision=revision,
-            subfolders=subfolders,
-        )
-        self.device = resolve_device(device)
-        self.batch_size = batch_size
-        self.max_length = max_length
-        self.cache_dir = cache_dir
-        self.local_files_only = local_files_only
-
-    def _load_kwargs(self, spec: ModelSpec) -> dict:
-        values = {
-            "revision": spec.revision,
-            "cache_dir": self.cache_dir,
-            "local_files_only": self.local_files_only,
-            "trust_remote_code": True,
-        }
-        if spec.subfolder is not None:
-            values["subfolder"] = spec.subfolder
-        return values
-
-    @staticmethod
-    def _validate_local_model_path(spec: ModelSpec) -> None:
-        path = Path(spec.model_id).expanduser()
-        config_path = path / "config.json"
-        if not path.is_dir() or not config_path.is_file():
-            return
-        try:
-            config_text = config_path.read_text()
-        except OSError:
-            return
-        if "DOEJGI/GenomeOcean-100M-v1.2" in config_text:
-            raise RuntimeError(
-                f"Model checkpoint {path} still references remote DOEJGI custom code. "
-                "Use a prepared Hugging Face inference directory instead of the raw "
-                "training checkpoint."
-            )
-
-    def _predict_one_model(
-        self,
-        spec: ModelSpec,
-        chunks: Sequence[SequenceChunk],
-    ) -> torch.Tensor:
-        self._validate_local_model_path(spec)
-        common = self._load_kwargs(spec)
-        tokenizer = AutoTokenizer.from_pretrained(
-            spec.model_id,
-            model_max_length=self.max_length,
-            use_fast=True,
-            padding_side="right",
-            **common,
-        )
-        model = AutoModelForSequenceClassification.from_pretrained(
-            spec.model_id,
-            **common,
-        )
-        model.to(self.device)
-        model.eval()
-
-        batches: list[torch.Tensor] = []
-        for start in range(0, len(chunks), self.batch_size):
-            batch = chunks[start : start + self.batch_size]
-            encoded = tokenizer(
-                [chunk.sequence for chunk in batch],
-                padding=True,
-                truncation=True,
-                max_length=self.max_length,
-                return_token_type_ids=False,
-                return_tensors="pt",
-            )
-            encoded.pop("token_type_ids", None)
-            encoded = {name: value.to(self.device) for name, value in encoded.items()}
-            with torch.inference_mode():
-                logits = model(**encoded).logits
-                probabilities = torch.softmax(logits, dim=-1).detach().cpu()
-            if probabilities.shape[1] != len(SUB_LABELS):
-                raise RuntimeError(
-                    f"Sub model '{spec.name}' returned {probabilities.shape[1]} labels; "
-                    f"expected {len(SUB_LABELS)}"
-                )
-            batches.append(probabilities)
-        return torch.cat(batches, dim=0)
-
-    def _release_device_cache(self) -> None:
-        gc.collect()
-        if self.device.type == "cuda":
-            torch.cuda.empty_cache()
-        elif self.device.type == "mps" and hasattr(torch.mps, "empty_cache"):
-            torch.mps.empty_cache()
-
-    def predict_chunks(self, chunks: Sequence[SequenceChunk]) -> pd.DataFrame:
-        if not chunks:
-            return pd.DataFrame(columns=SUB_CHUNK_RESULT_COLUMNS)
-
-        accumulator = ProbabilityEnsembleAccumulator(
-            n_items=len(chunks),
-            n_labels=len(SUB_LABELS),
-        )
-        for spec in self.model_specs:
-            try:
-                probabilities = self._predict_one_model(spec, chunks)
-                accumulator.add(probabilities)
-                del probabilities
-            finally:
-                self._release_device_cache()
-
-        mean, predicted, votes, agreement, confidence_std = accumulator.finalize()
-        rows: list[dict] = []
-        for index, chunk in enumerate(chunks):
-            label = int(predicted[index].item())
-            probability_row = mean[index]
-            row = chunk.to_dict()
-            row.pop("sequence")
-            row["predicted_label"] = label
-            row["predicted_name"] = SUB_LABELS[label]
-            row["confidence"] = float(probability_row[label].item())
-            row["ensemble_size"] = accumulator.n_models
-            row["ensemble_votes"] = int(votes[index].item())
-            row["ensemble_agreement"] = float(agreement[index].item())
-            row["confidence_std"] = float(confidence_std[index].item())
-            for label_index, column in PROBABILITY_COLUMNS.items():
-                row[column] = float(probability_row[label_index].item())
-            rows.append(row)
-        return pd.DataFrame(rows, columns=SUB_CHUNK_RESULT_COLUMNS)
-
-    def predict_records(
-        self,
-        records: Sequence[FastaRecord],
-        *,
-        chunk_size: int = DEFAULT_CHUNK_SIZE,
-        stride: int = DEFAULT_STRIDE,
-    ) -> SubPredictionBundle:
-        chunks, skipped = preprocess_records(
-            records,
-            chunk_size=chunk_size,
-            stride=stride,
-        )
-        chunk_results = self.predict_chunks(chunks)
-        contig_results = aggregate_chunks_to_contigs(chunk_results)
-        return SubPredictionBundle(
-            records=list(records),
-            chunks=chunks,
-            skipped=skipped,
-            chunk_results=chunk_results,
-            contig_results=contig_results,
-            file_results=aggregate_contigs_to_files(contig_results),
-        )
-
-    def predict_fasta(
-        self,
-        input_path: str | Path,
-        output_dir: str | Path,
-        *,
-        chunk_size: int = DEFAULT_CHUNK_SIZE,
-        stride: int = DEFAULT_STRIDE,
-    ) -> SubPredictionBundle:
-        records = read_input_records(input_path)
-        bundle = self.predict_records(records, chunk_size=chunk_size, stride=stride)
-        write_sub_outputs(
-            bundle,
-            output_dir,
-            metadata={
-                "input": str(Path(input_path).expanduser().resolve()),
-                "ensemble_method": ENSEMBLE_METHOD,
-                "ensemble_size": len(self.model_specs),
-                "models": [spec.to_dict() for spec in self.model_specs],
-                "sequential_model_loading": True,
-                "device": str(self.device),
-                "batch_size": self.batch_size,
-                "max_length": self.max_length,
-                "chunk_size": chunk_size,
-                "stride": stride,
-            },
-        )
-        return bundle
-
-
-def write_sub_outputs(
-    bundle: SubPredictionBundle,
-    output_dir: str | Path,
-    *,
-    metadata: dict | None = None,
-) -> dict[str, Path]:
-    out = Path(output_dir)
-    out.mkdir(parents=True, exist_ok=True)
-    paths = {
-        "chunks": out / "chunk_predictions.tsv",
-        "contigs": out / "contig_predictions.tsv",
-        "files": out / "file_predictions.tsv",
-        "skipped": out / "skipped_records.tsv",
-        "metadata": out / "run_metadata.json",
-    }
-    bundle.chunk_results.reindex(columns=SUB_CHUNK_RESULT_COLUMNS).to_csv(
-        paths["chunks"], sep="\t", index=False
-    )
-    bundle.contig_results.to_csv(paths["contigs"], sep="\t", index=False)
-    bundle.file_results.to_csv(paths["files"], sep="\t", index=False)
-    pd.DataFrame(
-        [item.to_dict() for item in bundle.skipped],
-        columns=SKIPPED_RESULT_COLUMNS,
-    ).to_csv(
-        paths["skipped"],
-        sep="\t",
-        index=False,
-    )
-    run_metadata = {
-        **(metadata or {}),
-        "records": len(bundle.records),
-        "chunks": len(bundle.chunks),
-        "predicted_contigs": len(bundle.contig_results),
-        "skipped_records": len(bundle.skipped),
-        "outputs": {name: str(path) for name, path in paths.items() if name != "metadata"},
-    }
-    paths["metadata"].write_text(json.dumps(run_metadata, indent=2) + "\n")
-    return paths

src/genomeocean_sub/preprocessing.py

@@ -1,119 +0,0 @@
-"""Training-compatible DNA cleaning and fixed-size chunking for Sub."""
-
-from __future__ import annotations
-
-from dataclasses import asdict, dataclass
-import re
-from typing import Iterable
-
-from .fasta import FastaRecord
-
-
-DEFAULT_CHUNK_SIZE = 5000
-DEFAULT_STRIDE = 5000
-
-
-@dataclass(frozen=True)
-class SequenceChunk:
-    record_id: str
-    source_file: str
-    contig_id: str
-    original_length: int
-    clean_length: int
-    n_chunks: int
-    ignored_tail_bp: int
-    chunk_index: int
-    chunk_start: int
-    chunk_end: int
-    sequence: str
-
-    def to_dict(self) -> dict:
-        return asdict(self)
-
-
-@dataclass(frozen=True)
-class SkippedRecord:
-    record_id: str
-    source_file: str
-    contig_id: str
-    original_length: int
-    clean_length: int
-    reason: str
-
-    def to_dict(self) -> dict:
-        return asdict(self)
-
-
-def clean_dna(sequence: str) -> str:
-    cleaned = re.sub(r"[^ACGTNacgtn]", "", sequence).upper()
-    return cleaned.replace("N", "")
-
-
-def make_chunk_spans(
-    sequence_length: int,
-    *,
-    chunk_size: int = DEFAULT_CHUNK_SIZE,
-    stride: int = DEFAULT_STRIDE,
-) -> list[tuple[int, int]]:
-    if chunk_size <= 0:
-        raise ValueError("chunk_size must be greater than zero")
-    if stride <= 0:
-        raise ValueError("stride must be greater than zero")
-    if sequence_length < chunk_size:
-        return []
-    return [
-        (start, start + chunk_size)
-        for start in range(0, sequence_length - chunk_size + 1, stride)
-    ]
-
-
-def preprocess_records(
-    records: Iterable[FastaRecord],
-    *,
-    chunk_size: int = DEFAULT_CHUNK_SIZE,
-    stride: int = DEFAULT_STRIDE,
-) -> tuple[list[SequenceChunk], list[SkippedRecord]]:
-    chunks: list[SequenceChunk] = []
-    skipped: list[SkippedRecord] = []
-    for record in records:
-        cleaned = clean_dna(record.sequence)
-        spans = make_chunk_spans(
-            len(cleaned),
-            chunk_size=chunk_size,
-            stride=stride,
-        )
-        if not spans:
-            skipped.append(
-                SkippedRecord(
-                    record_id=record.record_id,
-                    source_file=record.source_file,
-                    contig_id=record.contig_id,
-                    original_length=len(record.sequence),
-                    clean_length=len(cleaned),
-                    reason=(
-                        "empty_after_cleaning"
-                        if not cleaned
-                        else "shorter_than_chunk_size"
-                    ),
-                )
-            )
-            continue
-
-        ignored_tail_bp = max(0, len(cleaned) - max(end for _, end in spans))
-        for chunk_index, (start, end) in enumerate(spans):
-            chunks.append(
-                SequenceChunk(
-                    record_id=record.record_id,
-                    source_file=record.source_file,
-                    contig_id=record.contig_id,
-                    original_length=len(record.sequence),
-                    clean_length=len(cleaned),
-                    n_chunks=len(spans),
-                    ignored_tail_bp=ignored_tail_bp,
-                    chunk_index=chunk_index,
-                    chunk_start=start,
-                    chunk_end=end,
-                    sequence=cleaned[start:end],
-                )
-            )
-    return chunks, skipped

src/genomeocean_sub_classifier.egg-info/PKG-INFO

@@ -1,11 +0,0 @@
-Metadata-Version: 2.4
-Name: genomeocean-sub-classifier
-Version: 0.1.0
-Summary: GenomeOcean main classifier for genomic FASTA files
-Requires-Python: >=3.11
-Requires-Dist: torch<2.9,>=2.8
-Requires-Dist: transformers==4.51.3
-Requires-Dist: huggingface-hub<1,>=0.36
-Requires-Dist: safetensors<1,>=0.5
-Requires-Dist: numpy<3,>=2.2
-Requires-Dist: pandas<3,>=2.2

src/genomeocean_sub_classifier.egg-info/SOURCES.txt

@@ -1,16 +0,0 @@
-README.md
-pyproject.toml
-src/genomeocean_sub/__init__.py
-src/genomeocean_sub/aggregation.py
-src/genomeocean_sub/cli.py
-src/genomeocean_sub/fasta.py
-src/genomeocean_sub/predict.py
-src/genomeocean_sub/preprocessing.py
-src/genomeocean_sub_classifier.egg-info/PKG-INFO
-src/genomeocean_sub_classifier.egg-info/SOURCES.txt
-src/genomeocean_sub_classifier.egg-info/dependency_links.txt
-src/genomeocean_sub_classifier.egg-info/entry_points.txt
-src/genomeocean_sub_classifier.egg-info/requires.txt
-src/genomeocean_sub_classifier.egg-info/top_level.txt
-tests/test_preprocessing.py
-tests/test_smoke.py

src/genomeocean_sub_classifier.egg-info/dependency_links.txt

@@ -1 +0,0 @@
-

src/genomeocean_sub_classifier.egg-info/entry_points.txt

@@ -1,2 +0,0 @@
-[console_scripts]
-genomeocean-sub = genomeocean_sub.cli:main

src/genomeocean_sub_classifier.egg-info/requires.txt

@@ -1,6 +0,0 @@
-torch<2.9,>=2.8
-transformers==4.51.3
-huggingface-hub<1,>=0.36
-safetensors<1,>=0.5
-numpy<3,>=2.2
-pandas<3,>=2.2

src/genomeocean_sub_classifier.egg-info/top_level.txt

@@ -1 +0,0 @@
-genomeocean_sub

tests/test_preprocessing.py

@@ -1,28 +0,0 @@
-import unittest
-
-from genomeocean_sub.fasta import FastaRecord
-from genomeocean_sub.preprocessing import clean_dna, make_chunk_spans, preprocess_records
-
-
-class PreprocessingTests(unittest.TestCase):
-    def test_clean_dna_matches_training_behavior(self):
-        self.assertEqual(clean_dna("acgtn-RYSW"), "ACGT")
-
-    def test_exact_5kb_produces_one_chunk(self):
-        self.assertEqual(make_chunk_spans(5000), [(0, 5000)])
-
-    def test_short_record_is_reported(self):
-        record = FastaRecord("candidate.fna", "short", "short", "A" * 4999)
-        chunks, skipped = preprocess_records([record])
-        self.assertEqual(chunks, [])
-        self.assertEqual(skipped[0].reason, "shorter_than_chunk_size")
-
-    def test_main_and_sub_chunk_contract_is_5kb_non_overlapping(self):
-        self.assertEqual(
-            make_chunk_spans(10000),
-            [(0, 5000), (5000, 10000)],
-        )
-
-
-if __name__ == "__main__":
-    unittest.main()

tests/test_smoke.py

@@ -1,84 +0,0 @@
-import unittest
-from pathlib import Path
-
-import pandas as pd
-import torch
-
-from genomeocean_sub.aggregation import (
-    aggregate_chunks_to_contigs,
-    aggregate_contigs_to_files,
-)
-from genomeocean_sub.fasta import FastaRecord
-from genomeocean_sub.predict import (
-    ModelSpec,
-    ProbabilityEnsembleAccumulator,
-    SubPredictor,
-    build_model_specs,
-)
-from genomeocean_sub.preprocessing import preprocess_records
-
-
-class SmokeTests(unittest.TestCase):
-    def test_fold_specs_and_probability_mean(self):
-        specs = build_model_specs(
-            ["/models/sub-fold1", "/models/sub-fold2"],
-        )
-        self.assertEqual([spec.model_id for spec in specs], ["/models/sub-fold1", "/models/sub-fold2"])
-        self.assertEqual([spec.name for spec in specs], ["model1", "model2"])
-
-        accumulator = ProbabilityEnsembleAccumulator(n_items=1, n_labels=2)
-        accumulator.add(torch.tensor([[0.8, 0.2]]))
-        accumulator.add(torch.tensor([[0.4, 0.6]]))
-        mean, predicted, votes, agreement, _ = accumulator.finalize()
-        self.assertTrue(
-            torch.allclose(mean, torch.tensor([[0.6, 0.4]], dtype=torch.float64))
-        )
-        self.assertEqual(predicted.tolist(), [0])
-        self.assertEqual(votes.tolist(), [1])
-        self.assertEqual(agreement.tolist(), [0.5])
-
-    def test_raw_training_checkpoint_gets_export_hint(self):
-        checkpoint = (
-            Path(__file__).resolve().parents[4]
-            / "finetuning_go_sub_add"
-            / "ft_models"
-            / "train_100M_v1.2_5kb"
-            / "fold1"
-            / "checkpoint-44240"
-        )
-        if checkpoint.is_dir():
-            with self.assertRaisesRegex(RuntimeError, "prepared Hugging Face inference"):
-                SubPredictor._validate_local_model_path(ModelSpec(str(checkpoint)))
-
-    def test_fake_mirus_predictions_aggregate_to_mirus(self):
-        record = FastaRecord(
-            "candidates.fna",
-            "candidate_1",
-            "candidate_1",
-            "A" * 10000,
-        )
-        chunks, skipped = preprocess_records([record])
-        self.assertEqual(skipped, [])
-
-        rows = []
-        for chunk in chunks:
-            row = chunk.to_dict()
-            row.pop("sequence")
-            row.update(
-                {
-                    "predicted_label": 1,
-                    "predicted_name": "Mirus",
-                    "confidence": 0.9,
-                    "prob_ncldv": 0.1,
-                    "prob_mirus": 0.9,
-                }
-            )
-            rows.append(row)
-        contigs = aggregate_chunks_to_contigs(pd.DataFrame(rows))
-        files = aggregate_contigs_to_files(contigs)
-        self.assertEqual(contigs.loc[0, "predicted_name"], "Mirus")
-        self.assertEqual(files.loc[0, "predicted_name"], "Mirus")
-
-
-if __name__ == "__main__":
-    unittest.main()