├── .gitignore
├── LICENSE
├── README.md
├── app.py
├── chat.py
├── data
└── ultrachat_small.jsonl
├── requirements.txt
├── scripts
└── download_ultrachat.py
├── train_mamba.py
└── trainer
├── __init__.py
├── data.py
└── mamba_trainer.py
/.gitignore:
--------------------------------------------------------------------------------
1 | __pycache__/
2 | .pytest_cache/
3 | .venv/
4 | .vscode/
5 |
--------------------------------------------------------------------------------
/LICENSE:
--------------------------------------------------------------------------------
1 | Apache License
2 | Version 2.0, January 2004
3 | http://www.apache.org/licenses/
4 |
5 | TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
6 |
7 | 1. Definitions.
8 |
9 | "License" shall mean the terms and conditions for use, reproduction,
10 | and distribution as defined by Sections 1 through 9 of this document.
11 |
12 | "Licensor" shall mean the copyright owner or entity authorized by
13 | the copyright owner that is granting the License.
14 |
15 | "Legal Entity" shall mean the union of the acting entity and all
16 | other entities that control, are controlled by, or are under common
17 | control with that entity. For the purposes of this definition,
18 | "control" means (i) the power, direct or indirect, to cause the
19 | direction or management of such entity, whether by contract or
20 | otherwise, or (ii) ownership of fifty percent (50%) or more of the
21 | outstanding shares, or (iii) beneficial ownership of such entity.
22 |
23 | "You" (or "Your") shall mean an individual or Legal Entity
24 | exercising permissions granted by this License.
25 |
26 | "Source" form shall mean the preferred form for making modifications,
27 | including but not limited to software source code, documentation
28 | source, and configuration files.
29 |
30 | "Object" form shall mean any form resulting from mechanical
31 | transformation or translation of a Source form, including but
32 | not limited to compiled object code, generated documentation,
33 | and conversions to other media types.
34 |
35 | "Work" shall mean the work of authorship, whether in Source or
36 | Object form, made available under the License, as indicated by a
37 | copyright notice that is included in or attached to the work
38 | (an example is provided in the Appendix below).
39 |
40 | "Derivative Works" shall mean any work, whether in Source or Object
41 | form, that is based on (or derived from) the Work and for which the
42 | editorial revisions, annotations, elaborations, or other modifications
43 | represent, as a whole, an original work of authorship. For the purposes
44 | of this License, Derivative Works shall not include works that remain
45 | separable from, or merely link (or bind by name) to the interfaces of,
46 | the Work and Derivative Works thereof.
47 |
48 | "Contribution" shall mean any work of authorship, including
49 | the original version of the Work and any modifications or additions
50 | to that Work or Derivative Works thereof, that is intentionally
51 | submitted to Licensor for inclusion in the Work by the copyright owner
52 | or by an individual or Legal Entity authorized to submit on behalf of
53 | the copyright owner. For the purposes of this definition, "submitted"
54 | means any form of electronic, verbal, or written communication sent
55 | to the Licensor or its representatives, including but not limited to
56 | communication on electronic mailing lists, source code control systems,
57 | and issue tracking systems that are managed by, or on behalf of, the
58 | Licensor for the purpose of discussing and improving the Work, but
59 | excluding communication that is conspicuously marked or otherwise
60 | designated in writing by the copyright owner as "Not a Contribution."
61 |
62 | "Contributor" shall mean Licensor and any individual or Legal Entity
63 | on behalf of whom a Contribution has been received by Licensor and
64 | subsequently incorporated within the Work.
65 |
66 | 2. Grant of Copyright License. Subject to the terms and conditions of
67 | this License, each Contributor hereby grants to You a perpetual,
68 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable
69 | copyright license to reproduce, prepare Derivative Works of,
70 | publicly display, publicly perform, sublicense, and distribute the
71 | Work and such Derivative Works in Source or Object form.
72 |
73 | 3. Grant of Patent License. Subject to the terms and conditions of
74 | this License, each Contributor hereby grants to You a perpetual,
75 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable
76 | (except as stated in this section) patent license to make, have made,
77 | use, offer to sell, sell, import, and otherwise transfer the Work,
78 | where such license applies only to those patent claims licensable
79 | by such Contributor that are necessarily infringed by their
80 | Contribution(s) alone or by combination of their Contribution(s)
81 | with the Work to which such Contribution(s) was submitted. If You
82 | institute patent litigation against any entity (including a
83 | cross-claim or counterclaim in a lawsuit) alleging that the Work
84 | or a Contribution incorporated within the Work constitutes direct
85 | or contributory patent infringement, then any patent licenses
86 | granted to You under this License for that Work shall terminate
87 | as of the date such litigation is filed.
88 |
89 | 4. Redistribution. You may reproduce and distribute copies of the
90 | Work or Derivative Works thereof in any medium, with or without
91 | modifications, and in Source or Object form, provided that You
92 | meet the following conditions:
93 |
94 | (a) You must give any other recipients of the Work or
95 | Derivative Works a copy of this License; and
96 |
97 | (b) You must cause any modified files to carry prominent notices
98 | stating that You changed the files; and
99 |
100 | (c) You must retain, in the Source form of any Derivative Works
101 | that You distribute, all copyright, patent, trademark, and
102 | attribution notices from the Source form of the Work,
103 | excluding those notices that do not pertain to any part of
104 | the Derivative Works; and
105 |
106 | (d) If the Work includes a "NOTICE" text file as part of its
107 | distribution, then any Derivative Works that You distribute must
108 | include a readable copy of the attribution notices contained
109 | within such NOTICE file, excluding those notices that do not
110 | pertain to any part of the Derivative Works, in at least one
111 | of the following places: within a NOTICE text file distributed
112 | as part of the Derivative Works; within the Source form or
113 | documentation, if provided along with the Derivative Works; or,
114 | within a display generated by the Derivative Works, if and
115 | wherever such third-party notices normally appear. The contents
116 | of the NOTICE file are for informational purposes only and
117 | do not modify the License. You may add Your own attribution
118 | notices within Derivative Works that You distribute, alongside
119 | or as an addendum to the NOTICE text from the Work, provided
120 | that such additional attribution notices cannot be construed
121 | as modifying the License.
122 |
123 | You may add Your own copyright statement to Your modifications and
124 | may provide additional or different license terms and conditions
125 | for use, reproduction, or distribution of Your modifications, or
126 | for any such Derivative Works as a whole, provided Your use,
127 | reproduction, and distribution of the Work otherwise complies with
128 | the conditions stated in this License.
129 |
130 | 5. Submission of Contributions. Unless You explicitly state otherwise,
131 | any Contribution intentionally submitted for inclusion in the Work
132 | by You to the Licensor shall be under the terms and conditions of
133 | this License, without any additional terms or conditions.
134 | Notwithstanding the above, nothing herein shall supersede or modify
135 | the terms of any separate license agreement you may have executed
136 | with Licensor regarding such Contributions.
137 |
138 | 6. Trademarks. This License does not grant permission to use the trade
139 | names, trademarks, service marks, or product names of the Licensor,
140 | except as required for reasonable and customary use in describing the
141 | origin of the Work and reproducing the content of the NOTICE file.
142 |
143 | 7. Disclaimer of Warranty. Unless required by applicable law or
144 | agreed to in writing, Licensor provides the Work (and each
145 | Contributor provides its Contributions) on an "AS IS" BASIS,
146 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
147 | implied, including, without limitation, any warranties or conditions
148 | of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
149 | PARTICULAR PURPOSE. You are solely responsible for determining the
150 | appropriateness of using or redistributing the Work and assume any
151 | risks associated with Your exercise of permissions under this License.
152 |
153 | 8. Limitation of Liability. In no event and under no legal theory,
154 | whether in tort (including negligence), contract, or otherwise,
155 | unless required by applicable law (such as deliberate and grossly
156 | negligent acts) or agreed to in writing, shall any Contributor be
157 | liable to You for damages, including any direct, indirect, special,
158 | incidental, or consequential damages of any character arising as a
159 | result of this License or out of the use or inability to use the
160 | Work (including but not limited to damages for loss of goodwill,
161 | work stoppage, computer failure or malfunction, or any and all
162 | other commercial damages or losses), even if such Contributor
163 | has been advised of the possibility of such damages.
164 |
165 | 9. Accepting Warranty or Additional Liability. While redistributing
166 | the Work or Derivative Works thereof, You may choose to offer,
167 | and charge a fee for, acceptance of support, warranty, indemnity,
168 | or other liability obligations and/or rights consistent with this
169 | License. However, in accepting such obligations, You may act only
170 | on Your own behalf and on Your sole responsibility, not on behalf
171 | of any other Contributor, and only if You agree to indemnify,
172 | defend, and hold each Contributor harmless for any liability
173 | incurred by, or claims asserted against, such Contributor by reason
174 | of your accepting any such warranty or additional liability.
175 |
176 | END OF TERMS AND CONDITIONS
177 |
178 | APPENDIX: How to apply the Apache License to your work.
179 |
180 | To apply the Apache License to your work, attach the following
181 | boilerplate notice, with the fields enclosed by brackets "[]"
182 | replaced with your own identifying information. (Don't include
183 | the brackets!) The text should be enclosed in the appropriate
184 | comment syntax for the file format. We also recommend that a
185 | file or class name and description of purpose be included on the
186 | same "printed page" as the copyright notice for easier
187 | identification within third-party archives.
188 |
189 | Copyright [yyyy] [name of copyright owner]
190 |
191 | Licensed under the Apache License, Version 2.0 (the "License");
192 | you may not use this file except in compliance with the License.
193 | You may obtain a copy of the License at
194 |
195 | http://www.apache.org/licenses/LICENSE-2.0
196 |
197 | Unless required by applicable law or agreed to in writing, software
198 | distributed under the License is distributed on an "AS IS" BASIS,
199 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
200 | See the License for the specific language governing permissions and
201 | limitations under the License.
--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
1 | # Mamba-Chat 🐍
2 |
3 | **Mamba-Chat is the first chat language model based on a state-space model architecture, not a transformer.**
4 |
5 | The model is based on Albert Gu's and Tri Dao's work *Mamba: Linear-Time Sequence Modeling with Selective State Spaces* ([paper](https://arxiv.org/pdf/2312.00752.pdf)) as well as their [model implementation](https://github.com/state-spaces/mamba). This repository provides training / fine-tuning code for the model based on some modifications of the Huggingface Trainer class.
6 |
7 | Mamba-Chat is based on Mamba-2.8B and was fine-tuned on 16,000 samples of the [HuggingFaceH4/ultrachat_200k](https://huggingface.co/datasets/HuggingFaceH4/ultrachat_200k) dataset. To learn more, you can:
8 |
9 | - Take a look at the model on [Huggingface](https://huggingface.co/havenhq/mamba-chat) 🤗
10 | - Talk to us on the [Haven](https://haven.run/) Community [Discord](https://discord.com/invite/JDjbfp6q2G) 🧑🤝🧑
11 | - Talk to Mamba-Chat on [Google Colab](https://colab.research.google.com/drive/1dUlEYnRbgJYg4_kofNpsCddLCh6vltNK?usp=sharing)
12 |
13 |
14 |
15 |
16 | ## Run Mamba-Chat
17 |
18 | We provide code for testing and fine-tuning our model. Here's how to get started and what you can do with it:
19 |
20 |
21 |
22 |
23 | **Clone repository and install dependencies:**
24 | ```
25 | git clone https://github.com/havenhq/mamba-chat.git
26 | cd mamba-chat
27 | pip install -r requirements.txt
28 | ```
29 |
30 |
31 |
32 | **Talk to Mamba-Chat (CLI chatbot):**
33 | ```
34 | python chat.py
35 | ```
36 |
37 |
38 |
39 | **Talk to Mamba-Chat (gradio app):**
40 | ```
41 | pip install gradio==4.8.0
42 | python app.py --share
43 | ```
44 |
45 |
46 |
47 | **Fine-Tune Mamba (the base model) on a subset of the Ultrachat dataset:**
48 | ```
49 | python train_mamba.py --model state-spaces/mamba-2.8b --tokenizer EleutherAI/gpt-neox-20b --learning_rate 5e-5 --batch_size 4 --data_path ./data/ultrachat_small.jsonl --num_epochs 3
50 | ```
51 |
52 |
53 |
54 | **If you have a 24GB card (3090, 4090, etc.) you can use these settings:**
55 | ```
56 | python train_mamba.py --model state-spaces/mamba-2.8b --tokenizer EleutherAI/gpt-neox-20b --learning_rate 5e-5 --batch_size 1 --gradient_accumulation_steps 4 --optim paged_adamw_8bit --data_path ./data/ultrachat_small.jsonl --num_epochs 3
57 | ```
58 |
59 | ## Citation
60 |
61 | ```
62 | bibtex
63 | @misc{haven2023mambachat,
64 | title = {Mamba-Chat},
65 | author = {Justus Mattern and Konstantin Hohr},
66 | year = {2023},
67 | howpublished = {GitHub},
68 | url = {https://github.com/havenhq/mamba-chat}
69 | }
70 | ```
71 |
--------------------------------------------------------------------------------
/app.py:
--------------------------------------------------------------------------------
1 | import gradio as gr
2 | import torch
3 | from transformers import AutoTokenizer, AutoModelForCausalLM
4 | from mamba_ssm.models.mixer_seq_simple import MambaLMHeadModel
5 | from argparse import ArgumentParser
6 |
7 | def get_args():
8 | parser = ArgumentParser()
9 | parser.add_argument("--port", type=int, default=7860)
10 | parser.add_argument("--device", type=str, default='cuda', help='Device to run the model on')
11 | parser.add_argument("--model", type=str, default='havenhq/mamba-chat', help='Model to use')
12 | parser.add_argument(
13 | "--share",
14 | action="store_true",
15 | default=False,
16 | help="share your instance publicly through gradio",
17 | )
18 | try:
19 | args = parser.parse_args()
20 | except:
21 | parser.print_help()
22 | exit(0)
23 | return args
24 |
25 |
26 | if __name__ == "__main__":
27 | args = get_args()
28 |
29 | device = args.device
30 | model_name = args.model
31 | eos = "<|endoftext|>"
32 | tokenizer = AutoTokenizer.from_pretrained(model_name)
33 | tokenizer.eos_token = eos
34 | tokenizer.pad_token = tokenizer.eos_token
35 | tokenizer.chat_template = AutoTokenizer.from_pretrained(
36 | "HuggingFaceH4/zephyr-7b-beta"
37 | ).chat_template
38 |
39 | model = MambaLMHeadModel.from_pretrained(
40 | model_name, device=device, dtype=torch.float16
41 | )
42 |
43 | def chat_with_mamba(
44 | user_message,
45 | history: list[list[str]],
46 | temperature: float = 0.9,
47 | top_p: float = 0.7,
48 | max_length: int = 2000,
49 | ):
50 | history_dict: list[dict[str, str]] = []
51 | for user_m, assistant_m in history:
52 | history_dict.append(dict(role="user", content=user_m))
53 | history_dict.append(dict(role="assistant", content=assistant_m))
54 | history_dict.append(dict(role="user", content=user_message))
55 |
56 | input_ids = tokenizer.apply_chat_template(
57 | history_dict, return_tensors="pt", add_generation_prompt=True
58 | ).to(device)
59 |
60 | out = model.generate(
61 | input_ids=input_ids,
62 | max_length=max_length,
63 | temperature=temperature,
64 | top_p=top_p,
65 | eos_token_id=tokenizer.eos_token_id,
66 | )
67 |
68 | decoded = tokenizer.batch_decode(out)
69 | assistant_message = (
70 | decoded[0].split("<|assistant|>\n")[-1].replace(eos, "")
71 | )
72 | return assistant_message
73 |
74 |
75 | demo = gr.ChatInterface(
76 | fn=chat_with_mamba,
77 | # examples=[
78 | # "Explain what is state space model",
79 | # "Nice to meet you!",
80 | # "'Mamba is way better than ChatGPT.' Is this statement correct?",
81 | # ],
82 | additional_inputs=[
83 | gr.Slider(minimum=0, maximum=1, step=0.1, value=0.9, label="temperature"),
84 | gr.Slider(minimum=0, maximum=1, step=0.1, value=0.7, label="top_p"),
85 | gr.Number(value=2000, label="max_length"),
86 | ],
87 | title="Mamba Chat",
88 | )
89 | demo.launch(server_port=args.port, share=args.share)
90 |
--------------------------------------------------------------------------------
/chat.py:
--------------------------------------------------------------------------------
1 | import torch
2 | from transformers import AutoTokenizer, AutoModelForCausalLM
3 | from mamba_ssm.models.mixer_seq_simple import MambaLMHeadModel
4 |
5 | device = "cuda"
6 | tokenizer = AutoTokenizer.from_pretrained("havenhq/mamba-chat")
7 | tokenizer.eos_token = "<|endoftext|>"
8 | tokenizer.pad_token = tokenizer.eos_token
9 | tokenizer.chat_template = AutoTokenizer.from_pretrained("HuggingFaceH4/zephyr-7b-beta").chat_template
10 |
11 | model = MambaLMHeadModel.from_pretrained("havenhq/mamba-chat", device="cuda", dtype=torch.float16)
12 |
13 | messages = []
14 | while True:
15 | user_message = input("\nYour message: ")
16 | messages.append(dict(
17 | role="user",
18 | content=user_message
19 | ))
20 |
21 | input_ids = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True).to("cuda")
22 |
23 | out = model.generate(input_ids=input_ids, max_length=2000, temperature=0.9, top_p=0.7, eos_token_id=tokenizer.eos_token_id)
24 |
25 | decoded = tokenizer.batch_decode(out)
26 | messages.append(dict(
27 | role="assistant",
28 | content=decoded[0].split("<|assistant|>\n")[-1])
29 | )
30 |
31 | print("Model:", decoded[0].split("<|assistant|>\n")[-1])
--------------------------------------------------------------------------------
/requirements.txt:
--------------------------------------------------------------------------------
1 | packaging
2 | torch==2.1.0
3 | transformers==4.35.0
4 | causal-conv1d==1.0.0
5 | mamba-ssm==1.0.1
6 | accelerate==0.25.0
7 | bitsandbytes==0.41.3
8 | scipy==1.11.4
--------------------------------------------------------------------------------
/scripts/download_ultrachat.py:
--------------------------------------------------------------------------------
1 | import json
2 | from datasets import load_dataset
3 |
4 | data = load_dataset("HuggingFaceH4/ultrachat_200k", split="train_sft")
5 |
6 |
7 | with open("../data/ultrachat.jsonl", "w") as f:
8 | for d in data:
9 | f.write(json.dumps(dict(messages=d["messages"]))+"\n")
10 |
11 |
--------------------------------------------------------------------------------
/train_mamba.py:
--------------------------------------------------------------------------------
1 | import torch
2 | import argparse
3 |
4 | from mamba_ssm.models.mixer_seq_simple import MambaLMHeadModel
5 | from transformers import AutoTokenizer, TrainingArguments
6 | from trainer.data import ChatDataModule
7 | from trainer.mamba_trainer import MambaTrainer
8 |
9 |
10 | def run(args):
11 |
12 | model = MambaLMHeadModel.from_pretrained(args.model, dtype=torch.bfloat16, device="cuda")
13 |
14 | tokenizer = AutoTokenizer.from_pretrained(args.tokenizer)
15 | tokenizer.eos_token = "<|endoftext|>"
16 | tokenizer.pad_token = tokenizer.eos_token
17 | tokenizer.chat_template = AutoTokenizer.from_pretrained("HuggingFaceH4/zephyr-7b-beta").chat_template
18 |
19 |
20 | data_module = ChatDataModule(
21 | tokenizer=tokenizer,
22 | data_path=args.data_path,
23 | conversation_template=tokenizer.chat_template,
24 | max_tokens=2048
25 | )
26 |
27 |
28 | trainer = MambaTrainer(
29 | model=model,
30 | train_dataset=data_module.dataset,
31 | tokenizer=tokenizer,
32 | args=TrainingArguments(
33 | learning_rate=args.learning_rate,
34 | num_train_epochs=args.num_epochs,
35 | per_device_train_batch_size=args.batch_size,
36 | gradient_accumulation_steps=args.gradient_accumulation_steps,
37 | optim=args.optim,
38 | output_dir="mamba-chat",
39 | logging_steps=50,
40 | save_steps=500,
41 | ),
42 | data_collator=data_module.data_collator,
43 | )
44 |
45 | trainer.train()
46 |
47 |
48 | if __name__ == "__main__":
49 | parser = argparse.ArgumentParser()
50 | parser.add_argument("--model", type=str, default="state-spaces/mamba-2.8b")
51 | parser.add_argument("--tokenizer", type=str, default="EleutherAI/gpt-neox-20b")
52 | parser.add_argument("--learning_rate", type=float, default=5e-5)
53 | parser.add_argument("--batch_size", type=int, default=4)
54 | parser.add_argument("--gradient_accumulation_steps", type=int, default=1)
55 | parser.add_argument("--optim", type=str, default="adamw_torch")
56 | parser.add_argument("--data_path", type=str, default="./data/ultrachat_small.jsonl")
57 | parser.add_argument("--num_epochs", type=int, default=1)
58 | args = parser.parse_args()
59 |
60 | run(args)
61 |
--------------------------------------------------------------------------------
/trainer/__init__.py:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/redotvideo/mamba-chat/68e60823eb99d94d71d6c29cc203795a6312aea3/trainer/__init__.py
--------------------------------------------------------------------------------
/trainer/data.py:
--------------------------------------------------------------------------------
1 | import torch
2 | import transformers
3 | import json
4 |
5 | from dataclasses import dataclass
6 | from typing import Dict, Sequence
7 | from tqdm import tqdm
8 | from torch.utils.data import Dataset
9 |
10 |
11 | class ChatDataset(Dataset):
12 | def __init__(self, data_path: str, tokenizer: transformers.AutoTokenizer, conversation_template: str, max_tokens: int):
13 | super(ChatDataset, self).__init__()
14 | data = []
15 | with open(data_path, "r") as file:
16 | for line in file:
17 | try:
18 | data.append(json.loads(line))
19 | except Exception as e:
20 | print("json processing exception", e)
21 | continue
22 |
23 |
24 | data_dict = preprocess(data, tokenizer, conversation_template, max_tokens)
25 |
26 | self.input_ids = data_dict["input_ids"]
27 | self.labels = data_dict["labels"]
28 |
29 | def __len__(self):
30 | return len(self.input_ids)
31 |
32 | def __getitem__(self, i) -> Dict[str, torch.Tensor]:
33 | return dict(input_ids=self.input_ids[i], labels=self.labels[i])
34 |
35 |
36 | @dataclass
37 | class DataCollatorForChatDataset(object):
38 | """
39 | Collate examples for supervised fine-tuning.
40 | """
41 |
42 | tokenizer: transformers.PreTrainedTokenizer
43 |
44 | def __call__(self, instances: Sequence[Dict]) -> Dict[str, torch.Tensor]:
45 | input_ids, labels = tuple([instance[key] for instance in instances] for key in ("input_ids", "input_ids"))
46 | input_ids = torch.nn.utils.rnn.pad_sequence(input_ids, batch_first=True, padding_value=self.tokenizer.pad_token_id)
47 | labels = torch.nn.utils.rnn.pad_sequence(labels, batch_first=True, padding_value=-100)
48 |
49 | return dict(
50 | input_ids=input_ids,
51 | labels=labels,
52 | attention_mask=input_ids.ne(self.tokenizer.pad_token_id),
53 | )
54 |
55 |
56 | class ChatDataModule():
57 | def __init__(self, tokenizer: transformers.PreTrainedTokenizer, data_path: str, conversation_template, max_tokens: int):
58 |
59 | self.dataset = ChatDataset(tokenizer=tokenizer, data_path=data_path, conversation_template=conversation_template, max_tokens=max_tokens)
60 | self.data_collator = DataCollatorForChatDataset(tokenizer=tokenizer)
61 |
62 |
63 | def preprocess(conversations: Sequence[Sequence[dict]], tokenizer: transformers.PreTrainedTokenizer, conversation_template: str, max_tokens: int) -> Dict:
64 | """
65 | Preprocess the data by tokenizing.
66 | """
67 | all_input_ids = []
68 | all_label_ids = []
69 | tokenizer.use_default_system_prompt = False
70 |
71 | print("Tokenizing dataset...")
72 | for conv in tqdm(conversations):
73 | current_conv = conv["messages"]
74 | tokenized_responses = []
75 | for msg in current_conv:
76 | if msg["role"] == "assistant":
77 | tokenized_responses.append(tokenizer.encode(msg["content"], add_special_tokens=False))
78 |
79 | tokenized_conv = tokenizer.apply_chat_template(current_conv, chat_template=conversation_template, max_length=max_tokens, truncation=True)
80 | all_input_ids.append(torch.LongTensor(tokenized_conv))
81 |
82 |
83 | return dict(input_ids=all_input_ids, labels=all_input_ids)
--------------------------------------------------------------------------------
/trainer/mamba_trainer.py:
--------------------------------------------------------------------------------
1 | from transformers import Trainer
2 | import torch
3 | import os
4 |
5 |
6 | class MambaTrainer(Trainer):
7 | def compute_loss(self, model, inputs, return_outputs=False):
8 | input_ids = inputs.pop("input_ids")
9 | lm_logits = model(input_ids).logits
10 |
11 | labels = input_ids.to(lm_logits.device)
12 | shift_logits = lm_logits[:, :-1, :].contiguous()
13 | labels = labels[:, 1:].contiguous()
14 |
15 | loss_fct = torch.nn.CrossEntropyLoss()
16 | lm_loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), labels.view(-1))
17 |
18 | return lm_loss
19 |
20 | def save_model(self, output_dir, _internal_call):
21 | if not os.path.exists(output_dir):
22 | os.makedirs(output_dir)
23 |
24 | torch.save(self.model.state_dict(), f"{output_dir}/pytorch_model.bin")
25 | self.tokenizer.save_pretrained(output_dir)
--------------------------------------------------------------------------------