Mamba2 conversion script for original models (huggingface#32580)

* first attempt at allowing both conversions from codestral and from the original mamba ssm

* allow fp16, seems default for mamba2

* dtype fix

* simplify codestral check, dont overwrite pad/eos/bos when codestral

* change file -> directory

* use path join to be safe

* style

* apply code review
- add util mamba2 tokenizer (gptneox with left padding)
- add models dict

* fix copies

* add tokenizer to docs

* empty commit to check for weird err

* make conversion user dependent on model type, defaults for original paper models

* small comment nit

* remove norm_before_gate in conversion

* simplify model dict by using shared keys directly + remove unnecessary attributes

* fix tokenization: remove separate mamba2 tokenizer, add padding option as kwarg to gptneox one and reuse it for the conversion script

* simplify even further as we pass padding side via **kwargs already

src/transformers/models/mamba2/convert_mamba2_ssm_checkpoint_to_pytorch.py

@@ -15,55 +15,179 @@
 """This script can be used to convert checkpoints provided in the `mamba2_ssm` library into the format provided in HuggingFace `transformers`. It depends on the `mamba2_ssm` package to be installed."""
 
 import argparse
+import json
+from functools import partial
+from os import path
+from typing import Dict, Optional
 
 import torch
 from safetensors import safe_open
+from safetensors.torch import save_model
 
-from transformers import LlamaTokenizerFast, Mamba2Config, Mamba2ForCausalLM
+from transformers import GPTNeoXTokenizerFast, LlamaTokenizerFast, Mamba2Config, Mamba2ForCausalLM
 
 
-def convert_mamba2_checkpoint_file_to_huggingface_model_file(
-    mamba2_checkpoint_path: str, tokenizer_model_path: str, output_dir: str
-) -> None:
-    hf_config = Mamba2Config()
-    hf_model = Mamba2ForCausalLM(hf_config)
+def load_state_dict_from_safetensors(mamba2_checkpoint_path: str, ckpt_name: str) -> Dict[str, torch.Tensor]:
     # Load weights and config from paths
     original_state_dict = {}
-    with safe_open(mamba2_checkpoint_path, framework="pt") as f:
+    with safe_open(path.join(mamba2_checkpoint_path, ckpt_name), framework="pt") as f:
         for k in f.keys():
             newk = k.removeprefix("model.")
             original_state_dict[newk] = f.get_tensor(k).clone()
+    return original_state_dict
+
+
+def load_state_dict_from_torch(mamba2_checkpoint_path: str, ckpt_name: str) -> Dict[str, torch.Tensor]:
+    return torch.load(path.join(mamba2_checkpoint_path, ckpt_name), map_location="cpu")
+
+
+def convert_ssm_config_to_hf_config(config_ssm: Dict, mamba2_model_dict: Dict) -> Mamba2Config:
+    """Convert a Mamba2Config from mamba_ssm to a Mamba2Config from here."""
+    hf_config = Mamba2Config()
+
+    # Switch to a different dict depending on model type
+    config_dict = mamba2_model_dict
+
+    # Set important values from config and recalculate other resulting entries
+    hf_config.hidden_size = config_ssm[config_dict["hidden_size"]]
+    hf_config.num_heads = (hf_config.hidden_size * hf_config.expand) // hf_config.head_dim
+    hf_config.num_hidden_layers = config_ssm[config_dict["num_hidden_layers"]]
+    hf_config.n_groups = config_ssm.get(config_dict["n_groups"], 1)
+    hf_config.tie_word_embeddings = config_ssm["tie_embeddings"]
+    hf_config.bos_token_id = config_dict["bos_token_id"]
+    hf_config.pad_token_id = config_dict["pad_token_id"]
+    hf_config.eos_token_id = config_dict["eos_token_id"]
+
+    # Padded vocab size, mostly of 16 but 32 is also very common in different models
+    vocab_size = config_ssm["vocab_size"]
+    pad_vocab_size_multiple = config_ssm["pad_vocab_size_multiple"]
+    if (vocab_size % pad_vocab_size_multiple) != 0:
+        vocab_size += pad_vocab_size_multiple - (vocab_size % pad_vocab_size_multiple)
+    hf_config.vocab_size = vocab_size
+
+    return hf_config
+
+
+def load_and_save_tokenizer(
+    mamba2_model_type: str,
+    output_dir: str,
+    tokenizer_model_path: Optional[str] = None,
+) -> None:
+    tokenizer = None
+
+    # Load tokenizer
+    if tokenizer_model_path is not None and mamba2_model_type == "codestral":
+        tokenizer_class = LlamaTokenizerFast
+        tokenizer = tokenizer_class(tokenizer_model_path, legacy=False, from_slow=True)
+    elif mamba2_model_type == "mamba_ssm":
+        tokenizer = GPTNeoXTokenizerFast.from_pretrained("state-spaces/mamba-130m-hf", padding_side="left")
+
+    # Save tokenizer
+    if tokenizer is not None:
+        tokenizer.save_pretrained(output_dir)
 
+
+_MAMBA2_MODELS_DICT = {
+    "codestral": {
+        "hidden_size": "dim",
+        "num_hidden_layers": "n_layers",
+        "n_groups": "n_groups",
+        "bos_token_id": 0,
+        "pad_token_id": 1,
+        "eos_token_id": 2,
+        "config_name": "params.json",
+        "load_state_dict": partial(load_state_dict_from_safetensors, ckpt_name="consolidated.safetensors"),
+        "load_and_save_tokenizer": partial(load_and_save_tokenizer, "codestral"),
+    },
+    "mamba_ssm": {
+        "hidden_size": "d_model",
+        "num_hidden_layers": "n_layer",
+        "n_groups": "ngroups",
+        "bos_token_id": 0,
+        "pad_token_id": 0,
+        "eos_token_id": 0,
+        "config_name": "config.json",
+        "load_state_dict": partial(load_state_dict_from_torch, ckpt_name="pytorch_model.bin"),
+        "load_and_save_tokenizer": partial(load_and_save_tokenizer, "mamba_ssm"),
+    },
+}
+
+
+def convert_mamba2_checkpoint_file_to_huggingface_model_file(
+    mamba2_checkpoint_path: str,
+    mamba2_model_type: str,
+    precision: str,
+    output_dir: str,
+    tokenizer_model_path: Optional[str] = None,
+) -> None:
+    mamba2_model_dict = _MAMBA2_MODELS_DICT[mamba2_model_type]
+
+    # Load and save config based on name
+    config_path = path.join(mamba2_checkpoint_path, mamba2_model_dict["config_name"])
+    with open(config_path, "r", encoding="utf-8") as json_file:
+        config = json.load(json_file)
+    hf_config = convert_ssm_config_to_hf_config(config_ssm=config, mamba2_model_dict=mamba2_model_dict)
+    hf_config.save_pretrained(output_dir)
+
+    # Load state dict of the original model and transfer to hf model
+    original_state_dict = mamba2_model_dict["load_state_dict"](mamba2_checkpoint_path=mamba2_checkpoint_path)
+    hf_model = Mamba2ForCausalLM(hf_config)
     hf_model.load_state_dict(original_state_dict)
 
     # Save new model to pytorch_dump_path
-    hf_model.to(torch.bfloat16).save_pretrained(output_dir)
-    tokenizer_class = LlamaTokenizerFast
-    tokenizer = tokenizer_class(tokenizer_model_path, legacy=False, from_slow=True)
-    tokenizer.save_pretrained(output_dir)
+    dtype = torch.float32 if precision == "fp32" else (torch.bfloat16 if precision == "bf16" else torch.float16)
+    save_model(hf_model.to(dtype), path.join(output_dir, "model.safetensors"), metadata={"format": "pt"})
+
+    # Load and save tokenizer
+    mamba2_model_dict["load_and_save_tokenizer"](output_dir=output_dir, tokenizer_model_path=tokenizer_model_path)
 
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
     parser.add_argument(
         "-i",
-        "--mamba2_checkpoint_file",
+        "--mamba2_checkpoint_directory",
         type=str,
         required=True,
-        help="Path to a `pytorch_model.bin` mamba2_ssm checkpoint file to be converted.",
+        help="Path to a directory containing the `pytorch_model.bin` or `.safetensors` mamba2_ssm checkpoint file to be converted.",
     )
     parser.add_argument(
-        "-c",
-        "--tokenizer_model_path",
+        "-m",
+        "--mamba2_model_type",
+        type=str,
+        default="mamba_ssm",
+        const="mamba_ssm",
+        required=True,
+        choices=("codestral", "mamba_ssm"),
+        help="The model type the conversion will be performed on. Can choose from either `codestral` or `mamba_ssm`.",
+    )
+    parser.add_argument(
+        "-p",
+        "--precision",
         type=str,
+        default="fp16",
+        const="fp16",
         required=True,
-        help="Path to a `config.json` file corresponding to a Mamba2Config of the original mamba2_ssm model.",
+        choices=("fp32", "fp16", "bf16"),
+        help="The precision the model will be saved in. Select from fp32, fp16 or bf16.",
     )
     parser.add_argument(
         "-o", "--output_dir", type=str, required=True, help="Path to directory to save the converted output model to."
     )
+    parser.add_argument(
+        "-t",
+        "--tokenizer_model_path",
+        type=str,
+        default=None,
+        required=False,
+        help="Path to a `codestral` tokenizer file.",
+    )
     args = parser.parse_args()
 
     convert_mamba2_checkpoint_file_to_huggingface_model_file(
-        args.mamba2_checkpoint_file, args.tokenizer_model_path, args.output_dir
+        args.mamba2_checkpoint_directory,
+        args.mamba2_model_type,
+        args.precision,
+        args.output_dir,
+        args.tokenizer_model_path,
     )