Upload folder using huggingface_hub

config.json

@@ -1,105 +1,330 @@
 {
+  "transformers_version": "5.16.0.dev0",
   "architectures": [
-    "ESMFold2Model"
+    "EsmFold2Model"
   ],
-  "confidence_head": {
-    "distogram_bins": 39,
-    "enabled": true,
-    "folding_trunk": {
-      "dropout": 0.25,
-      "n_heads": 8,
-      "n_layers": 4
-    },
-    "max_dist": 50.75,
-    "min_dist": 3.25,
-    "num_pae_bins": 64,
-    "num_pde_bins": 64,
-    "num_plddt_bins": 50
-  },
-  "d_pair": 256,
-  "d_single": 384,
-  "disable_msa_features": false,
+  "output_hidden_states": false,
+  "return_dict": true,
   "dtype": "float32",
-  "esmc_id": "biohub/ESMC-6B",
-  "folding_trunk": {
-    "dropout": 0.25,
-    "n_heads": 8,
-    "n_layers": 48
-  },
-  "force_lm_dropout_during_inference": false,
-  "inputs": {
-    "atom_encoder": {
-      "d_atom": 128,
-      "d_token": 768,
-      "expansion_ratio": 2,
-      "n_blocks": 3,
-      "n_heads": 4,
-      "n_spatial_rope_pairs_per_axis": 2,
-      "n_uid_rope_pairs": 10,
-      "spatial_rope_base_frequency": 20.0,
-      "swa_window_size": 128,
-      "uid_rope_base_frequency": 10000.0
-    },
-    "d_inputs": 451
+  "chunk_size_feed_forward": 0,
+  "is_encoder_decoder": false,
+  "id2label": {
+    "0": "LABEL_0",
+    "1": "LABEL_1"
   },
-  "lm_d_model": 2560,
-  "lm_dropout": 0.0,
-  "lm_encoder": {
-    "enabled": true,
-    "lm_dropout": 0.25,
-    "n_layers": 4,
-    "per_loop_lm_dropout": true
-  },
-  "lm_num_layers": 80,
-  "model_type": "esmfold2",
-  "msa_encoder": {
-    "d_hidden": 32,
-    "d_msa": 128,
-    "enabled": true,
-    "msa_head_width": 16,
-    "n_heads_msa": 8,
-    "n_layers": 4
+  "label2id": {
+    "LABEL_0": 0,
+    "LABEL_1": 1
   },
-  "msa_encoder_overwrite": true,
-  "n_relative_chain_bins": 2,
-  "n_relative_residx_bins": 32,
-  "num_diffusion_samples": 32,
-  "num_loops": 3,
-  "parcae": {
-    "coda_n_layers": 2,
-    "enabled": true,
-    "max_steps": 6,
-    "min_steps": 1,
-    "poisson_mean": 3.0
+  "problem_type": null,
+  "hidden_size": 384,
+  "pairwise_hidden_size": 256,
+  "single_inputs_size": 451,
+  "pair_transition_intermediate_size": 1024,
+  "sliding_window": 128,
+  "chunk_size": 64,
+  "num_relative_residx_bins": 32,
+  "num_relative_chain_bins": 2,
+  "num_loops": 20,
+  "num_res_types": 33,
+  "max_atomic_number": 128,
+  "char_vocab_size": 64,
+  "max_chars": 4,
+  "max_atoms_per_token": 23,
+  "atom_feature_dim": 389,
+  "folding_trunk_num_hidden_layers": 48,
+  "parcae_num_coda_layers": 2,
+  "atom_encoder": {
+    "architectures": null,
+    "output_hidden_states": false,
+    "return_dict": true,
+    "dtype": null,
+    "chunk_size_feed_forward": 0,
+    "is_encoder_decoder": false,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "problem_type": null,
+    "hidden_size": 128,
+    "output_dim": 384,
+    "num_hidden_layers": 3,
+    "num_attention_heads": 4,
+    "head_dim": 32,
+    "intermediate_size": 256,
+    "spatial_rope_base_frequency": 20.0,
+    "num_spatial_rope_pairs_per_axis": 2,
+    "num_uid_rope_pairs": 10,
+    "uid_rope_base_frequency": 10000.0,
+    "_name_or_path": "",
+    "model_type": "",
+    "output_attentions": false,
+    "n_spatial_rope_pairs_per_axis": 2,
+    "n_uid_rope_pairs": 10,
+    "expansion_ratio": 2,
+    "token_hidden_size": 768
   },
   "structure_head": {
+    "architectures": null,
+    "output_hidden_states": false,
+    "return_dict": true,
+    "dtype": null,
+    "chunk_size_feed_forward": 0,
+    "is_encoder_decoder": false,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "problem_type": null,
     "diffusion_module": {
-      "atom_num_blocks": 3,
-      "atom_num_heads": 4,
-      "c_atom": 128,
-      "c_s_inputs": 451,
-      "c_token": 768,
-      "c_z": 256,
-      "fourier_dim": 256,
-      "relpos_r_max": 32,
-      "relpos_s_max": 2,
+      "architectures": null,
+      "output_hidden_states": false,
+      "return_dict": true,
+      "dtype": null,
+      "chunk_size_feed_forward": 0,
+      "is_encoder_decoder": false,
+      "id2label": {
+        "0": "LABEL_0",
+        "1": "LABEL_1"
+      },
+      "label2id": {
+        "LABEL_0": 0,
+        "LABEL_1": 1
+      },
+      "problem_type": null,
       "sigma_data": 16.0,
+      "hidden_size": 768,
+      "fourier_dim": 256,
+      "num_hidden_layers": 12,
+      "num_attention_heads": 16,
+      "head_dim": 48,
+      "intermediate_size": 1536,
+      "pair_intermediate_size": 512,
+      "atom_encoder": {
+        "architectures": null,
+        "output_hidden_states": false,
+        "return_dict": true,
+        "dtype": null,
+        "chunk_size_feed_forward": 0,
+        "is_encoder_decoder": false,
+        "id2label": {
+          "0": "LABEL_0",
+          "1": "LABEL_1"
+        },
+        "label2id": {
+          "LABEL_0": 0,
+          "LABEL_1": 1
+        },
+        "problem_type": null,
+        "hidden_size": 128,
+        "output_dim": 768,
+        "num_hidden_layers": 3,
+        "num_attention_heads": 4,
+        "head_dim": 32,
+        "intermediate_size": 256,
+        "spatial_rope_base_frequency": 20.0,
+        "num_spatial_rope_pairs_per_axis": 2,
+        "num_uid_rope_pairs": 10,
+        "uid_rope_base_frequency": 10000.0,
+        "_name_or_path": "",
+        "model_type": "",
+        "output_attentions": false
+      },
+      "_name_or_path": "",
+      "model_type": "",
+      "output_attentions": false,
+      "token_hidden_size": 768,
       "token_num_blocks": 12,
       "token_num_heads": 16,
-      "transition_multiplier": 2
+      "transition_multiplier": 2,
+      "c_z": 256,
+      "c_s_inputs": 451,
+      "relpos_r_max": 32,
+      "relpos_s_max": 2
     },
-    "distogram_bins": 64,
+    "num_distogram_bins": 64,
+    "num_diffusion_samples": 32,
     "gamma_0": 0.8,
     "gamma_min": 1.0,
+    "noise_scale": 1.003,
+    "step_scale": 1.5,
+    "inference_sigma_max_ratio": 160.0,
+    "inference_sigma_min_ratio": 0.0004,
+    "inference_exponent": 7.0,
     "inference_num_steps": 14,
-    "inference_p": 7.0,
+    "inference_sigma_cap": 256.0,
+    "_name_or_path": "",
+    "model_type": "",
+    "output_attentions": false,
+    "distogram_bins": 64,
     "inference_s_max": 160.0,
     "inference_s_min": 0.0004,
-    "noise_scale": 1.003,
-    "step_scale": 1.5,
+    "inference_p": 7.0,
     "train_noise_log_mean": -1.2,
     "train_noise_log_std": 1.5
   },
-  "transformers_version": "4.57.6",
-  "type": "release"
-}
+  "confidence_head": {
+    "architectures": null,
+    "output_hidden_states": false,
+    "return_dict": true,
+    "dtype": null,
+    "chunk_size_feed_forward": 0,
+    "is_encoder_decoder": false,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "problem_type": null,
+    "num_hidden_layers": 4,
+    "num_plddt_bins": 50,
+    "num_pde_bins": 64,
+    "num_pae_bins": 64,
+    "min_dist": 3.25,
+    "max_dist": 50.75,
+    "distogram_bins": 39,
+    "eps": 1e-06,
+    "_name_or_path": "",
+    "model_type": "",
+    "output_attentions": false,
+    "enabled": true,
+    "num_attention_heads": 8,
+    "dropout": 0.25
+  },
+  "msa_encoder": {
+    "architectures": null,
+    "output_hidden_states": false,
+    "return_dict": true,
+    "dtype": null,
+    "chunk_size_feed_forward": 0,
+    "is_encoder_decoder": false,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "problem_type": null,
+    "hidden_size": 128,
+    "outer_hidden_size": 32,
+    "num_hidden_layers": 4,
+    "num_attention_heads": 8,
+    "head_dim": 16,
+    "intermediate_size": 512,
+    "outer_product_chunk_size": null,
+    "_name_or_path": "",
+    "model_type": "",
+    "output_attentions": false,
+    "head_width": 16,
+    "enabled": true,
+    "overwrite": true,
+    "max_depth": 1024,
+    "column_mask_rate": 0.1
+  },
+  "lm_encoder": {
+    "architectures": null,
+    "output_hidden_states": false,
+    "return_dict": true,
+    "dtype": null,
+    "chunk_size_feed_forward": 0,
+    "is_encoder_decoder": false,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "problem_type": null,
+    "num_hidden_layers": 4,
+    "lm_dropout": 0.25,
+    "per_loop_lm_dropout": true,
+    "_name_or_path": "",
+    "model_type": "",
+    "output_attentions": false,
+    "enabled": true
+  },
+  "esmc_config": {
+    "architectures": null,
+    "output_hidden_states": false,
+    "return_dict": true,
+    "dtype": "float32",
+    "chunk_size_feed_forward": 0,
+    "is_encoder_decoder": false,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "problem_type": null,
+    "vocab_size": 64,
+    "hidden_size": 2560,
+    "intermediate_size": 6912,
+    "num_hidden_layers": 80,
+    "num_attention_heads": 40,
+    "num_key_value_heads": 40,
+    "hidden_act": "silu",
+    "max_position_embeddings": 2048,
+    "initializer_range": 0.02,
+    "pad_token_id": 1,
+    "bos_token_id": 0,
+    "eos_token_id": 2,
+    "tie_word_embeddings": false,
+    "rope_parameters": {
+      "rope_theta": 10000.0,
+      "rope_type": "default"
+    },
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "mlp_bias": false,
+    "head_dim": 64,
+    "mask_token_id": 32,
+    "classifier_dropout": 0.1,
+    "_name_or_path": "",
+    "model_type": "esmc",
+    "output_attentions": false
+  },
+  "_name_or_path": "",
+  "model_type": "esmfold2",
+  "output_attentions": false,
+  "n_relative_residx_bins": 32,
+  "n_relative_chain_bins": 2,
+  "type": "release",
+  "folding_trunk_num_attention_heads": 8,
+  "folding_trunk_dropout": 0.25,
+  "disable_msa_features": false,
+  "lm_dropout": 0.0,
+  "force_lm_dropout_during_inference": false,
+  "lm_d_model": 2560,
+  "lm_num_layers": 80,
+  "lm_mask_pct": 0.0,
+  "num_diffusion_samples": 32,
+  "parcae": {
+    "enabled": true,
+    "poisson_mean": 3.0,
+    "min_steps": 1,
+    "max_steps": 6,
+    "coda_n_layers": 2
+  },
+  "auto_map": {
+    "AutoConfig": "modeling_esmfold2_remote.EsmFold2RemoteConfig",
+    "AutoModel": "modeling_esmfold2_remote.EsmFold2RemoteModel"
+  }
+}

model-00001-of-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a24f050e2a2f9746b43b435c53719afc575a374fa302731b6911356a3d8b5cac
+size 4955672788

model-00002-of-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1af1615f6f6d1b3c0a24a63d9052a0ecad5b3831d1513c1a5fea73870c1bec73
+size 4971194264

model-00003-of-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:357b55ec56192fd01a1cfc39a5260caf2242503ac7c554f614e43c7a469f0ced
+size 4997460576

model-00004-of-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f065aca1a13308bce41b78f728c28a0a29e3a9db6af531a74a258dbb06f8721b
+size 4979069192

model-00005-of-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e0c507dce5b2ddb22f77305607b4332506137850c5378e98e9729345264668b7
+size 4934556472

model-00006-of-00006.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:90015eb13bbeab54237b18d6748dbd47f4bfd108fdf45704d66e5c3a93f8df7c
+size 1481396616

modeling_esmfold2_remote.py

@@ -0,0 +1,80 @@
+"""Load this checkpoint with the ``esm`` package instead of the transformers port.
+
+Shipped inside each ``biohub/ESMFold2*`` repo and named by its ``config.json``
+``auto_map``, so ``AutoModel.from_pretrained(repo, trust_remote_code=True)``
+returns an ``esm.models.esmfold2`` model. Without the flag transformers keeps
+using its own port.
+
+Runs from the Hub module cache in the user's environment, so it imports nothing
+but ``transformers``, ``packaging`` and ``esm``.
+"""
+
+from typing import Any
+
+from packaging.version import Version
+from transformers.configuration_utils import PretrainedConfig
+from transformers.modeling_utils import PreTrainedModel
+
+#: Inclusive floor. Older esm cannot read the bundled single-checkpoint layout.
+MIN_ESM_VERSION = "3.4.1"
+
+#: Everything esm's from_pretrained consumes, directly or via resolve_model_dir.
+_ESM_KWARGS = frozenset(
+    {
+        "load_esmc",
+        "esmc_precision",
+        "device",
+        "dtype",
+        "revision",
+        "cache_dir",
+        "token",
+        "local_files_only",
+        "force_download",
+    }
+)
+
+
+def esmfold2_class() -> Any:
+    """Return esm's ``EsmFold2Model``, or raise ``ImportError`` naming the pip command."""
+    try:
+        import esm
+        from esm.models.esmfold2 import EsmFold2Model
+    except ImportError as exc:
+        raise ImportError(
+            f"trust_remote_code=True runs the esm package, which failed to "
+            f"import ({exc}). Install it with:\n\n"
+            f"    pip install 'esm>={MIN_ESM_VERSION}'\n\n"
+            "Or drop the flag to use the ESMFold2 port in transformers."
+        ) from exc
+    if Version(esm.__version__) < Version(MIN_ESM_VERSION):
+        raise ImportError(
+            f"esm {esm.__version__} is installed, but this checkpoint needs "
+            f"{MIN_ESM_VERSION} or newer. Upgrade with:\n\n"
+            f"    pip install --upgrade 'esm>={MIN_ESM_VERSION}'"
+        )
+    return EsmFold2Model
+
+
+class EsmFold2RemoteConfig(PretrainedConfig):
+    """Holds config.json verbatim. esm re-reads the file and builds its own config."""
+
+    model_type = "esmfold2"
+
+
+class EsmFold2RemoteModel(PreTrainedModel):
+    """Loader shim; ``from_pretrained`` returns an esm model, not one of these.
+
+    A ``PreTrainedModel`` subclass only because the Auto classes call
+    ``register_for_auto_class`` on what they load and check its ``config_class``.
+    """
+
+    config_class = EsmFold2RemoteConfig
+
+    @classmethod
+    def from_pretrained(cls, pretrained_model_name_or_path, *args, **kwargs):  # type: ignore[override]
+        # esm forwards leftover kwargs to resolve_model_dir, whose signature is
+        # fixed, so transformers' trust_remote_code / _from_auto would TypeError.
+        forwarded = {k: v for k, v in kwargs.items() if k in _ESM_KWARGS}
+        return esmfold2_class().from_pretrained(
+            pretrained_model_name_or_path, *args, **forwarded
+        )

tokenizer.json

@@ -0,0 +1,167 @@
+{
+  "version": "1.0",
+  "truncation": null,
+  "padding": null,
+  "added_tokens": [
+    {
+      "id": 0,
+      "content": "<cls>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 1,
+      "content": "<pad>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 2,
+      "content": "<eos>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 3,
+      "content": "<unk>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 31,
+      "content": "|",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 32,
+      "content": "<mask>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    }
+  ],
+  "normalizer": null,
+  "pre_tokenizer": null,
+  "post_processor": {
+    "type": "TemplateProcessing",
+    "single": [
+      {
+        "SpecialToken": {
+          "id": "<cls>",
+          "type_id": 0
+        }
+      },
+      {
+        "Sequence": {
+          "id": "A",
+          "type_id": 0
+        }
+      },
+      {
+        "SpecialToken": {
+          "id": "<eos>",
+          "type_id": 0
+        }
+      }
+    ],
+    "pair": [
+      {
+        "Sequence": {
+          "id": "A",
+          "type_id": 0
+        }
+      },
+      {
+        "Sequence": {
+          "id": "B",
+          "type_id": 1
+        }
+      }
+    ],
+    "special_tokens": {
+      "<cls>": {
+        "id": "<cls>",
+        "ids": [
+          0
+        ],
+        "tokens": [
+          "<cls>"
+        ]
+      },
+      "<eos>": {
+        "id": "<eos>",
+        "ids": [
+          2
+        ],
+        "tokens": [
+          "<eos>"
+        ]
+      }
+    }
+  },
+  "decoder": null,
+  "model": {
+    "type": "BPE",
+    "dropout": null,
+    "unk_token": "<unk>",
+    "continuing_subword_prefix": null,
+    "end_of_word_suffix": null,
+    "fuse_unk": false,
+    "byte_fallback": false,
+    "ignore_merges": false,
+    "vocab": {
+      "<cls>": 0,
+      "<pad>": 1,
+      "<eos>": 2,
+      "<unk>": 3,
+      "L": 4,
+      "A": 5,
+      "G": 6,
+      "V": 7,
+      "S": 8,
+      "E": 9,
+      "R": 10,
+      "T": 11,
+      "I": 12,
+      "D": 13,
+      "P": 14,
+      "K": 15,
+      "Q": 16,
+      "N": 17,
+      "F": 18,
+      "Y": 19,
+      "M": 20,
+      "H": 21,
+      "W": 22,
+      "C": 23,
+      "X": 24,
+      "B": 25,
+      "U": 26,
+      "Z": 27,
+      "O": 28,
+      ".": 29,
+      "-": 30,
+      "|": 31,
+      "<mask>": 32
+    },
+    "merges": []
+  }
+}

tokenizer_config.json

@@ -0,0 +1,13 @@
+{
+  "backend": "tokenizers",
+  "bos_token": "<cls>",
+  "chain_break_token": "|",
+  "cls_token": "<cls>",
+  "eos_token": "<eos>",
+  "extra_special_tokens": [],
+  "mask_token": "<mask>",
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "<pad>",
+  "tokenizer_class": "EsmcTokenizer",
+  "unk_token": "<unk>"
+}