初始化项目，由ModelHub XC社区提供模型
Model: zstanjj/HTML-Pruner-Phi-3.8B Source: Original Platform
This commit is contained in:
ModelHub XC
38
.gitattributes
vendored
Normal file
38
.gitattributes
vendored
Normal file
@@ -0,0 +1,38 @@
|
||||
*.7z filter=lfs diff=lfs merge=lfs -text
|
||||
*.arrow filter=lfs diff=lfs merge=lfs -text
|
||||
*.bin filter=lfs diff=lfs merge=lfs -text
|
||||
*.bin.* filter=lfs diff=lfs merge=lfs -text
|
||||
*.bz2 filter=lfs diff=lfs merge=lfs -text
|
||||
*.ftz filter=lfs diff=lfs merge=lfs -text
|
||||
*.gz filter=lfs diff=lfs merge=lfs -text
|
||||
*.h5 filter=lfs diff=lfs merge=lfs -text
|
||||
*.joblib filter=lfs diff=lfs merge=lfs -text
|
||||
*.lfs.* filter=lfs diff=lfs merge=lfs -text
|
||||
*.model filter=lfs diff=lfs merge=lfs -text
|
||||
*.msgpack filter=lfs diff=lfs merge=lfs -text
|
||||
*.onnx filter=lfs diff=lfs merge=lfs -text
|
||||
*.ot filter=lfs diff=lfs merge=lfs -text
|
||||
*.parquet filter=lfs diff=lfs merge=lfs -text
|
||||
*.pb filter=lfs diff=lfs merge=lfs -text
|
||||
*.pt filter=lfs diff=lfs merge=lfs -text
|
||||
*.pth filter=lfs diff=lfs merge=lfs -text
|
||||
*.rar filter=lfs diff=lfs merge=lfs -text
|
||||
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
||||
*.tar.* filter=lfs diff=lfs merge=lfs -text
|
||||
*.tflite filter=lfs diff=lfs merge=lfs -text
|
||||
*.tgz filter=lfs diff=lfs merge=lfs -text
|
||||
*.xz filter=lfs diff=lfs merge=lfs -text
|
||||
*.zip filter=lfs diff=lfs merge=lfs -text
|
||||
*.zstandard filter=lfs diff=lfs merge=lfs -text
|
||||
*.tfevents* filter=lfs diff=lfs merge=lfs -text
|
||||
*.db* filter=lfs diff=lfs merge=lfs -text
|
||||
*.ark* filter=lfs diff=lfs merge=lfs -text
|
||||
**/*ckpt*data* filter=lfs diff=lfs merge=lfs -text
|
||||
**/*ckpt*.meta filter=lfs diff=lfs merge=lfs -text
|
||||
**/*ckpt*.index filter=lfs diff=lfs merge=lfs -text
|
||||
*.safetensors filter=lfs diff=lfs merge=lfs -text
|
||||
*.ckpt filter=lfs diff=lfs merge=lfs -text
|
||||
*.gguf* filter=lfs diff=lfs merge=lfs -text
|
||||
*.ggml filter=lfs diff=lfs merge=lfs -text
|
||||
*.llamafile* filter=lfs diff=lfs merge=lfs -text
|
||||
*.pt2 filter=lfs diff=lfs merge=lfs -text
|
||||
249
README.md
Normal file
249
README.md
Normal file
@@ -0,0 +1,249 @@
|
||||
---
|
||||
language:
|
||||
- en
|
||||
library_name: transformers
|
||||
base_model: microsoft/Phi-3.5-mini-instruct
|
||||
license: apache-2.0
|
||||
---
|
||||
|
||||
|
||||
## Model Information
|
||||
|
||||
We release the HTML pruner model used in **HtmlRAG: HTML is Better Than Plain Text for Modeling Retrieval Results in RAG Systems**.
|
||||
|
||||
<p align="left">
|
||||
Useful links: 📝 <a href="https://arxiv.org/abs/2411.02959" target="_blank">Paper</a> • 🤗 <a href="https://huggingface.co/papers/2411.02959" target="_blank">Hugging Face</a> • 🧩 <a href="https://github.com/plageon/HtmlRAG" target="_blank">Github</a>
|
||||
</p>
|
||||
|
||||
We propose HtmlRAG, which uses HTML instead of plain text as the format of external knowledge in RAG systems. To tackle the long context brought by HTML, we propose **Lossless HTML Cleaning** and **Two-Step Block-Tree-Based HTML Pruning**.
|
||||
|
||||
- **Lossless HTML Cleaning**: This cleaning process just removes totally irrelevant contents and compress redundant structures, retaining all semantic information in the original HTML. The compressed HTML of lossless HTML cleaning is suitable for RAG systems that have long-context LLMs and are not willing to loss any information before generation.
|
||||
|
||||
- **Two-Step Block-Tree-Based HTML Pruning**: The block-tree-based HTML pruning consists of two steps, both of which are conducted on the block tree structure. The first pruning step uses a embedding model to calculate scores for blocks, while the second step uses a path generative model. The first step processes the result of lossless HTML cleaning, while the second step processes the result of the first pruning step.
|
||||
|
||||
|
||||
🌹 If you use this model, please ✨star our **[GitHub repository](https://github.com/plageon/HtmlRAG)** to support us. Your star means a lot!
|
||||
|
||||
## 📦 Installation
|
||||
|
||||
Install the package using pip:
|
||||
```bash
|
||||
pip install htmlrag
|
||||
```
|
||||
Or install the package from source:
|
||||
```bash
|
||||
pip install -e .
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 📖 User Guide
|
||||
|
||||
### 🧹 HTML Cleaning
|
||||
|
||||
```python
|
||||
from htmlrag import clean_html
|
||||
|
||||
question = "When was the bellagio in las vegas built?"
|
||||
html = """
|
||||
<html>
|
||||
<head>
|
||||
<h1>Bellagio Hotel in Las</h1>
|
||||
</head>
|
||||
<body>
|
||||
<p class="class0">The Bellagio is a luxury hotel and casino located on the Las Vegas Strip in Paradise, Nevada. It was built in 1998.</p>
|
||||
</body>
|
||||
<div>
|
||||
<div>
|
||||
<p>Some other text</p>
|
||||
<p>Some other text</p>
|
||||
</div>
|
||||
</div>
|
||||
<p class="class1"></p>
|
||||
<!-- Some comment -->
|
||||
<script type="text/javascript">
|
||||
document.write("Hello World!");
|
||||
</script>
|
||||
</html>
|
||||
"""
|
||||
|
||||
#. alternatively you can read html files and merge them
|
||||
# html_files=["/path/to/html/file1.html", "/path/to/html/file2.html"]
|
||||
# htmls=[open(file).read() for file in html_files]
|
||||
# html = "\n".join(htmls)
|
||||
|
||||
simplified_html = clean_html(html)
|
||||
print(simplified_html)
|
||||
|
||||
# <html>
|
||||
# <h1>Bellagio Hotel in Las</h1>
|
||||
# <p>The Bellagio is a luxury hotel and casino located on the Las Vegas Strip in Paradise, Nevada. It was built in 1998.</p>
|
||||
# <div>
|
||||
# <p>Some other text</p>
|
||||
# <p>Some other text</p>
|
||||
# </div>
|
||||
# </html>
|
||||
```
|
||||
|
||||
### 🔧 Configure Pruning Parameters
|
||||
|
||||
The example HTML document is rather a short one. Real-world HTML documents can be much longer and more complex. To handle such cases, we can configure the following parameters:
|
||||
```python
|
||||
# Maximum number of words in a node when constructing the block tree for pruning with the embedding model
|
||||
MAX_NODE_WORDS_EMBED = 10
|
||||
# MAX_NODE_WORDS_EMBED = 256 # a recommended setting for real-world HTML documents
|
||||
# Maximum number of tokens in the output HTML document pruned with the embedding model
|
||||
MAX_CONTEXT_WINDOW_EMBED = 60
|
||||
# MAX_CONTEXT_WINDOW_EMBED = 6144 # a recommended setting for real-world HTML documents
|
||||
# Maximum number of words in a node when constructing the block tree for pruning with the generative model
|
||||
MAX_NODE_WORDS_GEN = 5
|
||||
# MAX_NODE_WORDS_GEN = 128 # a recommended setting for real-world HTML documents
|
||||
# Maximum number of tokens in the output HTML document pruned with the generative model
|
||||
MAX_CONTEXT_WINDOW_GEN = 32
|
||||
# MAX_CONTEXT_WINDOW_GEN = 4096 # a recommended setting for real-world HTML documents
|
||||
```
|
||||
|
||||
|
||||
|
||||
### 🌲 Build Block Tree
|
||||
|
||||
```python
|
||||
from htmlrag import build_block_tree
|
||||
|
||||
block_tree, simplified_html = build_block_tree(simplified_html, max_node_words=MAX_NODE_WORDS_EMBED)
|
||||
# block_tree, simplified_html = build_block_tree(simplified_html, max_node_words=MAX_NODE_WORDS_GEN, zh_char=True) # for Chinese text
|
||||
for block in block_tree:
|
||||
print("Block Content: ", block[0])
|
||||
print("Block Path: ", block[1])
|
||||
print("Is Leaf: ", block[2])
|
||||
print("")
|
||||
|
||||
# Block Content: <h1>Bellagio Hotel in Las</h1>
|
||||
# Block Path: ['html', 'title']
|
||||
# Is Leaf: True
|
||||
#
|
||||
# Block Content: <div>
|
||||
# <p>Some other text</p>
|
||||
# <p>Some other text</p>
|
||||
# </div>
|
||||
# Block Path: ['html', 'div']
|
||||
# Is Leaf: True
|
||||
#
|
||||
# Block Content: <p>The Bellagio is a luxury hotel and casino located on the Las Vegas Strip in Paradise, Nevada. It was built in 1998.</p>
|
||||
# Block Path: ['html', 'p']
|
||||
# Is Leaf: True
|
||||
```
|
||||
|
||||
### ✂️ Prune HTML Blocks with Embedding Model
|
||||
|
||||
```python
|
||||
from htmlrag import EmbedHTMLPruner
|
||||
|
||||
embed_model="BAAI/bge-large-en"
|
||||
query_instruction_for_retrieval = "Instruct: Given a web search query, retrieve relevant passages that answer the query\nQuery: "
|
||||
embed_html_pruner = EmbedHTMLPruner(embed_model=embed_model, local_inference=True, query_instruction_for_retrieval = query_instruction_for_retrieval)
|
||||
# alternatively you can init a remote TEI model, refer to https://github.com/huggingface/text-embeddings-inference.
|
||||
# tei_endpoint="http://YOUR_TEI_ENDPOINT"
|
||||
# embed_html_pruner = EmbedHTMLPruner(embed_model=embed_model, local_inference=False, query_instruction_for_retrieval = query_instruction_for_retrieval, endpoint=tei_endpoint)
|
||||
block_rankings=embed_html_pruner.calculate_block_rankings(question, simplified_html, block_tree)
|
||||
print(block_rankings)
|
||||
|
||||
# [2, 0, 1]
|
||||
|
||||
#. alternatively you can use bm25 to rank the blocks
|
||||
from htmlrag import BM25HTMLPruner
|
||||
bm25_html_pruner = BM25HTMLPruner()
|
||||
block_rankings = bm25_html_pruner.calculate_block_rankings(question, simplified_html, block_tree)
|
||||
print(block_rankings)
|
||||
|
||||
# [2, 0, 1]
|
||||
|
||||
from transformers import AutoTokenizer
|
||||
|
||||
chat_tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.1-70B-Instruct")
|
||||
|
||||
pruned_html = embed_html_pruner.prune_HTML(simplified_html, block_tree, block_rankings, chat_tokenizer, MAX_CONTEXT_WINDOW_EMBED)
|
||||
print(pruned_html)
|
||||
|
||||
# <html>
|
||||
# <h1>Bellagio Hotel in Las</h1>
|
||||
# <p>The Bellagio is a luxury hotel and casino located on the Las Vegas Strip in Paradise, Nevada. It was built in 1998.</p>
|
||||
# </html>
|
||||
```
|
||||
|
||||
|
||||
### ✂️ Prune HTML Blocks with Generative Model
|
||||
|
||||
```python
|
||||
from htmlrag import GenHTMLPruner
|
||||
import torch
|
||||
|
||||
# construct a finer block tree
|
||||
block_tree, pruned_html = build_block_tree(pruned_html, max_node_words=MAX_NODE_WORDS_GEN)
|
||||
# block_tree, pruned_html = build_block_tree(pruned_html, max_node_words=MAX_NODE_WORDS_GEN, zh_char=True) # for Chinese text
|
||||
for block in block_tree:
|
||||
print("Block Content: ", block[0])
|
||||
print("Block Path: ", block[1])
|
||||
print("Is Leaf: ", block[2])
|
||||
print("")
|
||||
|
||||
# Block Content: <h1>Bellagio Hotel in Las</h1>
|
||||
# Block Path: ['html', 'title']
|
||||
# Is Leaf: True
|
||||
#
|
||||
# Block Content: <p>The Bellagio is a luxury hotel and casino located on the Las Vegas Strip in Paradise, Nevada. It was built in 1998.</p>
|
||||
# Block Path: ['html', 'p']
|
||||
# Is Leaf: True
|
||||
|
||||
ckpt_path = "zstanjj/HTML-Pruner-Phi-3.8B"
|
||||
if torch.cuda.is_available():
|
||||
device="cuda"
|
||||
else:
|
||||
device="cpu"
|
||||
gen_embed_pruner = GenHTMLPruner(gen_model=ckpt_path, device=device)
|
||||
block_rankings = gen_embed_pruner.calculate_block_rankings(question, pruned_html, block_tree)
|
||||
print(block_rankings)
|
||||
|
||||
# [1, 0]
|
||||
|
||||
pruned_html = gen_embed_pruner.prune_HTML(pruned_html, block_tree, block_rankings, chat_tokenizer, MAX_CONTEXT_WINDOW_GEN)
|
||||
print(pruned_html)
|
||||
|
||||
# <p>The Bellagio is a luxury hotel and casino located on the Las Vegas Strip in Paradise, Nevada. It was built in 1998.</p>
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
|
||||
## Results
|
||||
|
||||
- **Results for [HTML-Pruner-Phi-3.8B](https://huggingface.co/zstanjj/HTML-Pruner-Phi-3.8B) and [HTML-Pruner-Llama-1B](https://huggingface.co/zstanjj/HTML-Pruner-Llama-1B) with Llama-3.1-70B-Instruct as chat model**.
|
||||
|
||||
| Dataset | ASQA | HotpotQA | NQ | TriviaQA | MuSiQue | ELI5 |
|
||||
|------------------|-----------|-----------|-----------|-----------|-----------|-----------|
|
||||
| Metrics | EM | EM | EM | EM | EM | ROUGE-L |
|
||||
| BM25 | 49.50 | 38.25 | 47.00 | 88.00 | 9.50 | 16.15 |
|
||||
| BGE | 68.00 | 41.75 | 59.50 | 93.00 | 12.50 | 16.20 |
|
||||
| E5-Mistral | 63.00 | 36.75 | 59.50 | 90.75 | 11.00 | 16.17 |
|
||||
| LongLLMLingua | 62.50 | 45.00 | 56.75 | 92.50 | 10.25 | 15.84 |
|
||||
| JinaAI Reader | 55.25 | 34.25 | 48.25 | 90.00 | 9.25 | 16.06 |
|
||||
| HtmlRAG-Phi-3.8B | **68.50** | **46.25** | 60.50 | **93.50** | **13.25** | **16.33** |
|
||||
| HtmlRAG-Llama-1B | 66.50 | 45.00 | **60.75** | 93.00 | 10.00 | 16.25 |
|
||||
|
||||
---
|
||||
|
||||
## 📜 Citation
|
||||
|
||||
```bibtex
|
||||
@misc{tan2024htmlraghtmlbetterplain,
|
||||
title={HtmlRAG: HTML is Better Than Plain Text for Modeling Retrieved Knowledge in RAG Systems},
|
||||
author={Jiejun Tan and Zhicheng Dou and Wen Wang and Mang Wang and Weipeng Chen and Ji-Rong Wen},
|
||||
year={2024},
|
||||
eprint={2411.02959},
|
||||
archivePrefix={arXiv},
|
||||
primaryClass={cs.IR},
|
||||
url={https://arxiv.org/abs/2411.02959},
|
||||
}
|
||||
```
|
||||
|
||||
|
||||
13
added_tokens.json
Normal file
13
added_tokens.json
Normal file
@@ -0,0 +1,13 @@
|
||||
{
|
||||
"<|assistant|>": 32001,
|
||||
"<|endoftext|>": 32000,
|
||||
"<|end|>": 32007,
|
||||
"<|placeholder1|>": 32002,
|
||||
"<|placeholder2|>": 32003,
|
||||
"<|placeholder3|>": 32004,
|
||||
"<|placeholder4|>": 32005,
|
||||
"<|placeholder5|>": 32008,
|
||||
"<|placeholder6|>": 32009,
|
||||
"<|system|>": 32006,
|
||||
"<|user|>": 32010
|
||||
}
|
||||
140
config.json
Normal file
140
config.json
Normal file
@@ -0,0 +1,140 @@
|
||||
{
|
||||
"_name_or_path": "Phi-3.5-mini-instruct",
|
||||
"architectures": [
|
||||
"Phi3ForCausalLM"
|
||||
],
|
||||
"attention_dropout": 0.0,
|
||||
"auto_map": {
|
||||
"AutoConfig": "configuration_phi3.Phi3Config",
|
||||
"AutoModelForCausalLM": "modeling_phi3.Phi3ForCausalLM",
|
||||
"AutoModelForSeq2SeqLM": "modeling_phi3.PHI3ForHTMLTreeGeneration"
|
||||
},
|
||||
"bos_token_id": 1,
|
||||
"embd_pdrop": 0.0,
|
||||
"eos_token_id": 32000,
|
||||
"hidden_act": "silu",
|
||||
"hidden_size": 3072,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 8192,
|
||||
"max_position_embeddings": 131072,
|
||||
"model_type": "phi3",
|
||||
"num_attention_heads": 32,
|
||||
"num_hidden_layers": 32,
|
||||
"num_key_value_heads": 32,
|
||||
"original_max_position_embeddings": 4096,
|
||||
"pad_token_id": 32000,
|
||||
"resid_pdrop": 0.0,
|
||||
"rms_norm_eps": 1e-05,
|
||||
"rope_scaling": {
|
||||
"long_factor": [
|
||||
1.0800000429153442,
|
||||
1.1100000143051147,
|
||||
1.1399999856948853,
|
||||
1.340000033378601,
|
||||
1.5899999141693115,
|
||||
1.600000023841858,
|
||||
1.6200000047683716,
|
||||
2.620000123977661,
|
||||
3.2300000190734863,
|
||||
3.2300000190734863,
|
||||
4.789999961853027,
|
||||
7.400000095367432,
|
||||
7.700000286102295,
|
||||
9.09000015258789,
|
||||
12.199999809265137,
|
||||
17.670000076293945,
|
||||
24.46000099182129,
|
||||
28.57000160217285,
|
||||
30.420001983642578,
|
||||
30.840002059936523,
|
||||
32.590003967285156,
|
||||
32.93000411987305,
|
||||
42.320003509521484,
|
||||
44.96000289916992,
|
||||
50.340003967285156,
|
||||
50.45000457763672,
|
||||
57.55000305175781,
|
||||
57.93000411987305,
|
||||
58.21000289916992,
|
||||
60.1400032043457,
|
||||
62.61000442504883,
|
||||
62.62000274658203,
|
||||
62.71000289916992,
|
||||
63.1400032043457,
|
||||
63.1400032043457,
|
||||
63.77000427246094,
|
||||
63.93000411987305,
|
||||
63.96000289916992,
|
||||
63.970001220703125,
|
||||
64.02999877929688,
|
||||
64.06999969482422,
|
||||
64.08000183105469,
|
||||
64.12000274658203,
|
||||
64.41000366210938,
|
||||
64.4800033569336,
|
||||
64.51000213623047,
|
||||
64.52999877929688,
|
||||
64.83999633789062
|
||||
],
|
||||
"short_factor": [
|
||||
1.0,
|
||||
1.0199999809265137,
|
||||
1.0299999713897705,
|
||||
1.0299999713897705,
|
||||
1.0499999523162842,
|
||||
1.0499999523162842,
|
||||
1.0499999523162842,
|
||||
1.0499999523162842,
|
||||
1.0499999523162842,
|
||||
1.0699999332427979,
|
||||
1.0999999046325684,
|
||||
1.1099998950958252,
|
||||
1.1599998474121094,
|
||||
1.1599998474121094,
|
||||
1.1699998378753662,
|
||||
1.2899998426437378,
|
||||
1.339999794960022,
|
||||
1.679999828338623,
|
||||
1.7899998426437378,
|
||||
1.8199998140335083,
|
||||
1.8499997854232788,
|
||||
1.8799997568130493,
|
||||
1.9099997282028198,
|
||||
1.9399996995925903,
|
||||
1.9899996519088745,
|
||||
2.0199997425079346,
|
||||
2.0199997425079346,
|
||||
2.0199997425079346,
|
||||
2.0199997425079346,
|
||||
2.0199997425079346,
|
||||
2.0199997425079346,
|
||||
2.0299997329711914,
|
||||
2.0299997329711914,
|
||||
2.0299997329711914,
|
||||
2.0299997329711914,
|
||||
2.0299997329711914,
|
||||
2.0299997329711914,
|
||||
2.0299997329711914,
|
||||
2.0299997329711914,
|
||||
2.0299997329711914,
|
||||
2.0799996852874756,
|
||||
2.0899996757507324,
|
||||
2.189999580383301,
|
||||
2.2199995517730713,
|
||||
2.5899994373321533,
|
||||
2.729999542236328,
|
||||
2.749999523162842,
|
||||
2.8399994373321533
|
||||
],
|
||||
"type": "longrope"
|
||||
},
|
||||
"rope_theta": 10000.0,
|
||||
"sliding_window": 262144,
|
||||
"tie_word_embeddings": false,
|
||||
"torch_dtype": "bfloat16",
|
||||
"transformers_version": "4.43.3",
|
||||
"use_cache": true,
|
||||
"attention_bias": false,
|
||||
"vocab_size": 32064,
|
||||
"attn_implementation": "flash_attention_2"
|
||||
}
|
||||
1
configuration.json
Normal file
1
configuration.json
Normal file
@@ -0,0 +1 @@
|
||||
{"framework":"Pytorch","task":"text-ranking"}
|
||||
227
configuration_phi3.py
Normal file
227
configuration_phi3.py
Normal file
@@ -0,0 +1,227 @@
|
||||
# coding=utf-8
|
||||
# Copyright 2024 Microsoft and the HuggingFace Inc. team. All rights reserved.
|
||||
#
|
||||
# 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.
|
||||
|
||||
""" Phi-3 model configuration"""
|
||||
|
||||
|
||||
from transformers.configuration_utils import PretrainedConfig
|
||||
from transformers.utils import logging
|
||||
|
||||
|
||||
logger = logging.get_logger(__name__)
|
||||
|
||||
PHI3_PRETRAINED_CONFIG_ARCHIVE_MAP = {
|
||||
"microsoft/Phi-3-mini-4k-instruct": "https://huggingface.co/microsoft/Phi-3-mini-4k-instruct/resolve/main/config.json",
|
||||
"microsoft/Phi-3-mini-128k-instruct": "https://huggingface.co/microsoft/Phi-3-mini-128k-instruct/resolve/main/config.json",
|
||||
}
|
||||
|
||||
|
||||
class Phi3Config(PretrainedConfig):
|
||||
r"""
|
||||
This is the configuration class to store the configuration of a [`Phi3Model`]. It is used to instantiate a Phi-3
|
||||
model according to the specified arguments, defining the model architecture. Instantiating a configuration with the
|
||||
defaults will yield a similar configuration to that of the
|
||||
[microsoft/Phi-3-mini-4k-instruct](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct).
|
||||
|
||||
Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
|
||||
documentation from [`PretrainedConfig`] for more information.
|
||||
|
||||
Args:
|
||||
vocab_size (`int`, *optional*, defaults to 32064):
|
||||
Vocabulary size of the Phi-3 model. Defines the number of different tokens that can be represented by the
|
||||
`inputs_ids` passed when calling [`Phi3Model`].
|
||||
hidden_size (`int`, *optional*, defaults to 3072):
|
||||
Dimension of the hidden representations.
|
||||
intermediate_size (`int`, *optional*, defaults to 8192):
|
||||
Dimension of the MLP representations.
|
||||
num_hidden_layers (`int`, *optional*, defaults to 32):
|
||||
Number of hidden layers in the Transformer decoder.
|
||||
num_attention_heads (`int`, *optional*, defaults to 32):
|
||||
Number of attention heads for each attention layer in the Transformer decoder.
|
||||
num_key_value_heads (`int`, *optional*):
|
||||
This is the number of key_value heads that should be used to implement Grouped Query Attention. If
|
||||
`num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
|
||||
`num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When
|
||||
converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
|
||||
by meanpooling all the original heads within that group. For more details checkout [this
|
||||
paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to
|
||||
`num_attention_heads`.
|
||||
resid_pdrop (`float`, *optional*, defaults to 0.0):
|
||||
Dropout probability for mlp outputs.
|
||||
embd_pdrop (`int`, *optional*, defaults to 0.0):
|
||||
The dropout ratio for the embeddings.
|
||||
attention_dropout (`float`, *optional*, defaults to 0.0):
|
||||
The dropout ratio after computing the attention scores.
|
||||
hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
|
||||
The non-linear activation function (function or string) in the decoder.
|
||||
max_position_embeddings (`int`, *optional*, defaults to 4096):
|
||||
The maximum sequence length that this model might ever be used with.
|
||||
original_max_position_embeddings (`int`, *optional*, defaults to 4096):
|
||||
The maximum sequence length that this model was trained with. This is used to determine the size of the
|
||||
original RoPE embeddings when using long scaling.
|
||||
initializer_range (`float`, *optional*, defaults to 0.02):
|
||||
The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
|
||||
rms_norm_eps (`float`, *optional*, defaults to 1e-05):
|
||||
The epsilon value used for the RMSNorm.
|
||||
use_cache (`bool`, *optional*, defaults to `True`):
|
||||
Whether or not the model should return the last key/values attentions (not used by all models). Only
|
||||
relevant if `config.is_decoder=True`. Whether to tie weight embeddings or not.
|
||||
tie_word_embeddings (`bool`, *optional*, defaults to `False`):
|
||||
Whether to tie weight embeddings
|
||||
rope_theta (`float`, *optional*, defaults to 10000.0):
|
||||
The base period of the RoPE embeddings.
|
||||
rope_scaling (`dict`, *optional*):
|
||||
The scaling strategy for the RoPE embeddings. If `None`, no scaling is applied. If a dictionary, it must
|
||||
contain the following keys: `type`, `short_factor` and `long_factor`. The `type` must be `longrope` and
|
||||
the `short_factor` and `long_factor` must be lists of numbers with the same length as the hidden size
|
||||
divided by the number of attention heads divided by 2.
|
||||
bos_token_id (`int`, *optional*, defaults to 1):
|
||||
The id of the "beginning-of-sequence" token.
|
||||
eos_token_id (`int`, *optional*, defaults to 32000):
|
||||
The id of the "end-of-sequence" token.
|
||||
pad_token_id (`int`, *optional*, defaults to 32000):
|
||||
The id of the padding token.
|
||||
sliding_window (`int`, *optional*):
|
||||
Sliding window attention window size. If `None`, no sliding window is applied.
|
||||
|
||||
Example:
|
||||
|
||||
```python
|
||||
>>> from transformers import Phi3Model, Phi3Config
|
||||
|
||||
>>> # Initializing a Phi-3 style configuration
|
||||
>>> configuration = Phi3Config.from_pretrained("microsoft/Phi-3-mini-4k-instruct")
|
||||
|
||||
>>> # Initializing a model from the configuration
|
||||
>>> model = Phi3Model(configuration)
|
||||
|
||||
>>> # Accessing the model configuration
|
||||
>>> configuration = model.config
|
||||
```"""
|
||||
|
||||
model_type = "phi3"
|
||||
keys_to_ignore_at_inference = ["past_key_values"]
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
vocab_size=32064,
|
||||
hidden_size=3072,
|
||||
intermediate_size=8192,
|
||||
num_hidden_layers=32,
|
||||
num_attention_heads=32,
|
||||
num_key_value_heads=None,
|
||||
resid_pdrop=0.0,
|
||||
embd_pdrop=0.0,
|
||||
attention_dropout=0.0,
|
||||
hidden_act="silu",
|
||||
max_position_embeddings=4096,
|
||||
original_max_position_embeddings=4096,
|
||||
initializer_range=0.02,
|
||||
rms_norm_eps=1e-5,
|
||||
use_cache=True,
|
||||
tie_word_embeddings=False,
|
||||
rope_theta=10000.0,
|
||||
rope_scaling=None,
|
||||
bos_token_id=1,
|
||||
eos_token_id=32000,
|
||||
pad_token_id=32000,
|
||||
sliding_window=None,
|
||||
**kwargs,
|
||||
):
|
||||
self.vocab_size = vocab_size
|
||||
self.hidden_size = hidden_size
|
||||
self.intermediate_size = intermediate_size
|
||||
self.num_hidden_layers = num_hidden_layers
|
||||
self.num_attention_heads = num_attention_heads
|
||||
|
||||
if num_key_value_heads is None:
|
||||
num_key_value_heads = num_attention_heads
|
||||
|
||||
self.num_key_value_heads = num_key_value_heads
|
||||
self.resid_pdrop = resid_pdrop
|
||||
self.embd_pdrop = embd_pdrop
|
||||
self.attention_dropout = attention_dropout
|
||||
self.hidden_act = hidden_act
|
||||
self.max_position_embeddings = max_position_embeddings
|
||||
self.original_max_position_embeddings = original_max_position_embeddings
|
||||
self.initializer_range = initializer_range
|
||||
self.rms_norm_eps = rms_norm_eps
|
||||
self.use_cache = use_cache
|
||||
self.rope_theta = rope_theta
|
||||
self.rope_scaling = rope_scaling
|
||||
self._rope_scaling_adjustment()
|
||||
self._rope_scaling_validation()
|
||||
self.sliding_window = sliding_window
|
||||
|
||||
super().__init__(
|
||||
bos_token_id=bos_token_id,
|
||||
eos_token_id=eos_token_id,
|
||||
pad_token_id=pad_token_id,
|
||||
tie_word_embeddings=tie_word_embeddings,
|
||||
**kwargs,
|
||||
)
|
||||
|
||||
def _rope_scaling_adjustment(self):
|
||||
"""
|
||||
Adjust the `type` of the `rope_scaling` configuration for backward compatibility.
|
||||
"""
|
||||
if self.rope_scaling is None:
|
||||
return
|
||||
|
||||
rope_scaling_type = self.rope_scaling.get("type", None)
|
||||
|
||||
# For backward compatibility if previous version used "su" or "yarn"
|
||||
if rope_scaling_type is not None and rope_scaling_type in ["su", "yarn"]:
|
||||
self.rope_scaling["type"] = "longrope"
|
||||
|
||||
def _rope_scaling_validation(self):
|
||||
"""
|
||||
Validate the `rope_scaling` configuration.
|
||||
"""
|
||||
if self.rope_scaling is None:
|
||||
return
|
||||
|
||||
if not isinstance(self.rope_scaling, dict) or len(self.rope_scaling) != 3:
|
||||
raise ValueError(
|
||||
"`rope_scaling` must be a dictionary with three fields, `type`, `short_factor` and `long_factor`, "
|
||||
f"got {self.rope_scaling}"
|
||||
)
|
||||
rope_scaling_type = self.rope_scaling.get("type", None)
|
||||
rope_scaling_short_factor = self.rope_scaling.get("short_factor", None)
|
||||
rope_scaling_long_factor = self.rope_scaling.get("long_factor", None)
|
||||
if rope_scaling_type is None or rope_scaling_type not in ["longrope"]:
|
||||
raise ValueError(f"`rope_scaling`'s type field must be one of ['longrope'], got {rope_scaling_type}")
|
||||
if not (
|
||||
isinstance(rope_scaling_short_factor, list)
|
||||
and all(isinstance(x, (int, float)) for x in rope_scaling_short_factor)
|
||||
):
|
||||
raise ValueError(
|
||||
f"`rope_scaling`'s short_factor field must be a list of numbers, got {rope_scaling_short_factor}"
|
||||
)
|
||||
if not len(rope_scaling_short_factor) == self.hidden_size // self.num_attention_heads // 2:
|
||||
raise ValueError(
|
||||
f"`rope_scaling`'s short_factor field must have length {self.hidden_size // self.num_attention_heads // 2}, got {len(rope_scaling_short_factor)}"
|
||||
)
|
||||
if not (
|
||||
isinstance(rope_scaling_long_factor, list)
|
||||
and all(isinstance(x, (int, float)) for x in rope_scaling_long_factor)
|
||||
):
|
||||
raise ValueError(
|
||||
f"`rope_scaling`'s long_factor field must be a list of numbers, got {rope_scaling_long_factor}"
|
||||
)
|
||||
if not len(rope_scaling_long_factor) == self.hidden_size // self.num_attention_heads // 2:
|
||||
raise ValueError(
|
||||
f"`rope_scaling`'s long_factor field must have length {self.hidden_size // self.num_attention_heads // 2}, got {len(rope_scaling_long_factor)}"
|
||||
)
|
||||
11
generation_config.json
Normal file
11
generation_config.json
Normal file
@@ -0,0 +1,11 @@
|
||||
{
|
||||
"_from_model_config": true,
|
||||
"bos_token_id": 1,
|
||||
"eos_token_id": [
|
||||
32007,
|
||||
32001,
|
||||
32000
|
||||
],
|
||||
"pad_token_id": 32000,
|
||||
"transformers_version": "4.43.4"
|
||||
}
|
||||
3
model-00001-of-00002.safetensors
Normal file
3
model-00001-of-00002.safetensors
Normal file
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:4cc7fb73476017fd9fc8cc3e4060b0896a407145659833751b238422297caba5
|
||||
size 4972489328
|
||||
3
model-00002-of-00002.safetensors
Normal file
3
model-00002-of-00002.safetensors
Normal file
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:1b3be9f3c5028c619facb98ea12e1626b04cb24043a83858f728e96e6211a1e7
|
||||
size 2669692552
|
||||
202
model.safetensors.index.json
Normal file
202
model.safetensors.index.json
Normal file
@@ -0,0 +1,202 @@
|
||||
{
|
||||
"metadata": {
|
||||
"total_size": 7642159104
|
||||
},
|
||||
"weight_map": {
|
||||
"lm_head.weight": "model-00002-of-00002.safetensors",
|
||||
"model.embed_tokens.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.21.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.21.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.3.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.30.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.mlp.gate_up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.self_attn.qkv_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.4.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.mlp.gate_up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.self_attn.qkv_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.norm.weight": "model-00002-of-00002.safetensors"
|
||||
}
|
||||
}
|
||||
1931
modeling_phi3.py
Normal file
1931
modeling_phi3.py
Normal file
File diff suppressed because it is too large
Load Diff
164
seq_para_utils.py
Normal file
164
seq_para_utils.py
Normal file
@@ -0,0 +1,164 @@
|
||||
import os
|
||||
import torch
|
||||
import logging
|
||||
import transformers
|
||||
import torch.distributed as dist
|
||||
import torch
|
||||
import math
|
||||
|
||||
# global var
|
||||
_SEQUENCE_PARALLEL_GROUP = None
|
||||
_SEQUENCE_PARALLEL_SIZE = 1
|
||||
|
||||
def init_logger(fpath='', local_rank=0):
|
||||
if transformers.trainer_utils.is_main_process(local_rank):
|
||||
if fpath:
|
||||
if os.path.dirname(fpath):
|
||||
os.makedirs(os.path.dirname(fpath), exist_ok=True)
|
||||
file_handler = logging.FileHandler(fpath, mode='a') # to file
|
||||
transformers.logging.add_handler(file_handler)
|
||||
transformers.logging.set_verbosity_info()
|
||||
else:
|
||||
transformers.logging.set_verbosity_error() # reduce
|
||||
transformers.logging.enable_explicit_format()
|
||||
return transformers.logging.get_logger()
|
||||
|
||||
class DistributedSampler(torch.utils.data.distributed.DistributedSampler):
|
||||
def set_epoch(self, epoch):
|
||||
# 重载Sample 保证每个epoch dataset更新后sampler 重新更新
|
||||
# If the dataset length is evenly divisible by # of replicas, then there
|
||||
# is no need to drop any data, since the dataset will be split equally.
|
||||
if self.drop_last and len(self.dataset) % self.num_replicas != 0: # type: ignore[arg-type]
|
||||
# Split to nearest available length that is evenly divisible.
|
||||
# This is to ensure each rank receives the same amount of data when
|
||||
# using this Sampler.
|
||||
self.num_samples = math.ceil(
|
||||
(len(self.dataset) - self.num_replicas) / self.num_replicas # type: ignore[arg-type]
|
||||
)
|
||||
else:
|
||||
self.num_samples = math.ceil(len(self.dataset) / self.num_replicas) # type: ignore[arg-type]
|
||||
self.total_size = self.num_samples * self.num_replicas
|
||||
super().set_epoch(epoch)
|
||||
|
||||
def add_custom_callback(trainer, logger):
|
||||
if 'PrinterCallback' in trainer.callback_handler.callback_list:
|
||||
trainer.pop_callback(transformers.PrinterCallback)
|
||||
trainer.add_callback(LogCallback(logger))
|
||||
logger.info('Add custom LogCallback')
|
||||
trainer.add_callback(DatasetUpdateCallback(trainer))
|
||||
logger.info('Add custom DatasetUpdateCallback')
|
||||
trainer.add_callback(SaveDiskCallback())
|
||||
logger.info('Add custom SaveDiskCallback')
|
||||
logger.info(f"trainer's callbacks: {trainer.callback_handler.callback_list}")
|
||||
|
||||
|
||||
class LogCallback(transformers.TrainerCallback):
|
||||
"""
|
||||
A bare :class:`~transformers.TrainerCallback` that just prints with logger.
|
||||
"""
|
||||
def __init__(self, logger, exclude=('total_flos', 'epoch')):
|
||||
self.logger = logger
|
||||
self.exclude = exclude
|
||||
|
||||
def on_log(self, args, state, control, logs=None, **kwargs):
|
||||
if state.is_world_process_zero:
|
||||
self.logger.info(''.join([
|
||||
f"[global_steps={state.global_step}]",
|
||||
f"[epochs={logs['epoch']}]",
|
||||
','.join(f'{k}={v}' for k, v in logs.items()
|
||||
if k not in self.exclude)
|
||||
]))
|
||||
|
||||
|
||||
class DatasetUpdateCallback(transformers.TrainerCallback):
|
||||
def __init__(self, trainer):
|
||||
self.trainer = trainer
|
||||
|
||||
def on_epoch_begin(self, args, state, control,train_dataloader, **kwargs):
|
||||
self.trainer.train_dataset.update(int(state.epoch))
|
||||
train_dataloader.sampler.set_epoch(int(state.epoch))
|
||||
|
||||
|
||||
class SaveDiskCallback(transformers.TrainerCallback):
|
||||
def on_save(self, args, state, control, **kwargs):
|
||||
if args.local_rank != 0:
|
||||
return
|
||||
|
||||
for ckpt in os.listdir(args.output_dir):
|
||||
# remove out-of-date deepspeed checkpoints
|
||||
if ckpt.startswith('checkpoint-') and not ckpt.endswith(f'-{state.global_step}'):
|
||||
for pattern in ['global_step*', '*.pth']:
|
||||
os.system("rm -rf " + os.path.join(args.output_dir, ckpt, pattern))
|
||||
|
||||
def on_train_end(self, args, state, control, **kwargs):
|
||||
if state.is_local_process_zero and False:
|
||||
for pattern in ['global_step*', '*.pth']:
|
||||
os.system("rm -rf " + os.path.join(args.output_dir, "checkpoint-*", pattern))
|
||||
|
||||
|
||||
def register_nan_hook(model):
|
||||
torch.autograd.set_detect_anomaly(True)
|
||||
|
||||
def add_module_name(module):
|
||||
for name, sub_module in module.named_modules():
|
||||
sub_module.name = name
|
||||
|
||||
def add_check_nan_hook(module):
|
||||
def check_nan(module, inputs, outputs):
|
||||
any_nan = False
|
||||
for i, tensor in enumerate(inputs):
|
||||
if isinstance(tensor, torch.Tensor) and tensor.isnan().any():
|
||||
print(f'module {module.name} contains nan in its {i}th input.')
|
||||
any_nan = True
|
||||
for i, tensor in enumerate(outputs):
|
||||
if isinstance(tensor, torch.Tensor) and tensor.isnan().any():
|
||||
print(f'module {module.name} contains nan in its {i}th output.')
|
||||
any_nan = True
|
||||
if any_nan:
|
||||
if torch.distributed.get_rank() == 0:
|
||||
torch.save({
|
||||
'state_dict': module.state_dict(),
|
||||
'inputs': inputs,
|
||||
'outputs': outputs,
|
||||
'type': module.__class__.__name__
|
||||
}, module.name + '.pth')
|
||||
# from ipdb import set_trace; set_trace()
|
||||
# else:
|
||||
# import time; time.sleep(10000)
|
||||
|
||||
module.register_forward_hook(lambda module, inputs, outputs: check_nan(module, inputs, outputs))
|
||||
module.register_forward_hook(lambda module, inputs, outputs: check_nan(module, inputs, outputs))
|
||||
|
||||
model.apply(add_module_name)
|
||||
model.apply(add_check_nan_hook)
|
||||
|
||||
|
||||
def initialize_seq_parallel(
|
||||
sequence_parallel_size,
|
||||
):
|
||||
if sequence_parallel_size <= 1:
|
||||
return None
|
||||
num_sequence_parallel_groups: int = dist.get_world_size() // sequence_parallel_size
|
||||
global _SEQUENCE_PARALLEL_GROUP
|
||||
global _SEQUENCE_PARALLEL_SIZE
|
||||
_SEQUENCE_PARALLEL_SIZE = sequence_parallel_size
|
||||
for i in range(num_sequence_parallel_groups):
|
||||
ranks = range(i * sequence_parallel_size,
|
||||
(i + 1) * sequence_parallel_size)
|
||||
group = torch.distributed.new_group(ranks)
|
||||
if dist.get_rank() in ranks:
|
||||
_SEQUENCE_PARALLEL_GROUP = group
|
||||
|
||||
def get_sequence_parallel_group():
|
||||
"""Get the sequence parallel group the caller rank belongs to."""
|
||||
return _SEQUENCE_PARALLEL_GROUP
|
||||
|
||||
def get_sequence_parallel_size():
|
||||
return _SEQUENCE_PARALLEL_SIZE
|
||||
|
||||
def get_sequence_parallel_rank():
|
||||
return torch.distributed.get_rank(group=get_sequence_parallel_group())
|
||||
|
||||
# 设置序列并行参数来保证优化器正确平均
|
||||
from deepspeed.utils import groups
|
||||
groups._get_sequence_parallel_world_size = get_sequence_parallel_size
|
||||
30
special_tokens_map.json
Normal file
30
special_tokens_map.json
Normal file
@@ -0,0 +1,30 @@
|
||||
{
|
||||
"bos_token": {
|
||||
"content": "<s>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false
|
||||
},
|
||||
"eos_token": {
|
||||
"content": "<|endoftext|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false
|
||||
},
|
||||
"pad_token": {
|
||||
"content": "<|endoftext|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false
|
||||
},
|
||||
"unk_token": {
|
||||
"content": "<unk>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false
|
||||
}
|
||||
}
|
||||
3
tokenizer.model
Normal file
3
tokenizer.model
Normal file
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
|
||||
size 499723
|
||||
132
tokenizer_config.json
Normal file
132
tokenizer_config.json
Normal file
@@ -0,0 +1,132 @@
|
||||
{
|
||||
"add_bos_token": false,
|
||||
"add_eos_token": false,
|
||||
"add_prefix_space": true,
|
||||
"added_tokens_decoder": {
|
||||
"0": {
|
||||
"content": "<unk>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"1": {
|
||||
"content": "<s>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"2": {
|
||||
"content": "</s>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": true,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"32000": {
|
||||
"content": "<|endoftext|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"32001": {
|
||||
"content": "<|assistant|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": true,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"32002": {
|
||||
"content": "<|placeholder1|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": true,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"32003": {
|
||||
"content": "<|placeholder2|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": true,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"32004": {
|
||||
"content": "<|placeholder3|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": true,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"32005": {
|
||||
"content": "<|placeholder4|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": true,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"32006": {
|
||||
"content": "<|system|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": true,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"32007": {
|
||||
"content": "<|end|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": true,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"32008": {
|
||||
"content": "<|placeholder5|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": true,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"32009": {
|
||||
"content": "<|placeholder6|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": true,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"32010": {
|
||||
"content": "<|user|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": true,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
}
|
||||
},
|
||||
"bos_token": "<s>",
|
||||
"chat_template": "{% for message in messages %}{% if message['role'] == 'system' and message['content'] %}{{'<|system|>\n' + message['content'] + '<|end|>\n'}}{% elif message['role'] == 'user' %}{{'<|user|>\n' + message['content'] + '<|end|>\n'}}{% elif message['role'] == 'assistant' %}{{'<|assistant|>\n' + message['content'] + '<|end|>\n'}}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ '<|assistant|>\n' }}{% else %}{{ eos_token }}{% endif %}",
|
||||
"clean_up_tokenization_spaces": false,
|
||||
"eos_token": "<|endoftext|>",
|
||||
"legacy": false,
|
||||
"model_max_length": 35000,
|
||||
"pad_token": "<|endoftext|>",
|
||||
"padding_side": "left",
|
||||
"sp_model_kwargs": {},
|
||||
"spaces_between_special_tokens": false,
|
||||
"tokenizer_class": "LlamaTokenizer",
|
||||
"unk_token": "<unk>",
|
||||
"use_default_system_prompt": false
|
||||
}
|
||||
106
tree_gen_utils.py
Normal file
106
tree_gen_utils.py
Normal file
@@ -0,0 +1,106 @@
|
||||
from collections import defaultdict
|
||||
from typing import List, Tuple
|
||||
|
||||
import numpy as np
|
||||
from anytree import Node, RenderTree
|
||||
import bs4
|
||||
from anytree import PreOrderIter
|
||||
from anytree.exporter import DotExporter
|
||||
|
||||
|
||||
def nodenamefunc(node):
|
||||
return f"{node.name}|{node.prob}|{node.input_ids}"
|
||||
|
||||
|
||||
class TokenDotExporter(DotExporter):
|
||||
def __init__(self, node, **kwargs):
|
||||
super().__init__(node, **kwargs)
|
||||
|
||||
def __iter__(self):
|
||||
# prepare
|
||||
indent = " " * self.indent
|
||||
nodenamefunc = self.nodenamefunc or self._default_nodenamefunc
|
||||
nodeattrfunc = self.nodeattrfunc or self._default_nodeattrfunc
|
||||
edgeattrfunc = self.edgeattrfunc or self._default_edgeattrfunc
|
||||
edgetypefunc = self.edgetypefunc or self._default_edgetypefunc
|
||||
filter_ = self.filter_ or self._default_filter
|
||||
return self.__iter(indent, nodenamefunc, nodeattrfunc, edgeattrfunc, edgetypefunc, filter_)
|
||||
|
||||
def __iter_nodes(self, indent, nodenamefunc, nodeattrfunc, filter_):
|
||||
for node in PreOrderIter(self.node, filter_=filter_, stop=self.stop, maxlevel=self.maxlevel):
|
||||
nodename = nodenamefunc(node)
|
||||
nodeattr = nodeattrfunc(node)
|
||||
nodeattr = " {%s}" % nodeattr if nodeattr is not None else ""
|
||||
yield '%s%s' % (DotExporter.esc(nodename), nodeattr)
|
||||
|
||||
def __iter(self, indent, nodenamefunc, nodeattrfunc, edgeattrfunc, edgetypefunc, filter_):
|
||||
for node in self.__iter_nodes(indent, nodenamefunc, nodeattrfunc, filter_):
|
||||
yield node
|
||||
|
||||
|
||||
class TokenIdNode(Node):
|
||||
def __init__(self, name, parent=None, children=None, **kwargs):
|
||||
super().__init__(name, parent, children, **kwargs)
|
||||
self.input_ids = kwargs.get('input_ids', [])
|
||||
self.prob = kwargs.get('prob', np.float32(0.0))
|
||||
|
||||
|
||||
def split_tree(soup: bs4.BeautifulSoup, max_node_words=0) -> List[Tuple[bs4.element.Tag, List[str], bool]]:
|
||||
word_count = len(soup.get_text().split())
|
||||
if word_count > max_node_words:
|
||||
possible_trees = [(soup, [])]
|
||||
target_trees = [] # [(tag, path, is_leaf)]
|
||||
# split the entire dom tee into subtrees, until the length of the subtree is less than max_node_words words
|
||||
# find all possible trees
|
||||
while True:
|
||||
if len(possible_trees) == 0:
|
||||
break
|
||||
tree = possible_trees.pop(0)
|
||||
tag_children = defaultdict(int)
|
||||
bare_word_count = 0
|
||||
# count child tags
|
||||
for child in tree[0].contents:
|
||||
if isinstance(child, bs4.element.Tag):
|
||||
tag_children[child.name] += 1
|
||||
_tag_children = {k: 0 for k in tag_children.keys()}
|
||||
|
||||
# check if the tree can be split
|
||||
for child in tree[0].contents:
|
||||
if isinstance(child, bs4.element.Tag):
|
||||
# change child tag with duplicate names
|
||||
if tag_children[child.name] > 1:
|
||||
new_name = f"{child.name}{_tag_children[child.name]}"
|
||||
new_tree = (child, tree[1] + [new_name])
|
||||
_tag_children[child.name] += 1
|
||||
child.name = new_name
|
||||
else:
|
||||
new_tree = (child, tree[1] + [child.name])
|
||||
word_count = len(child.get_text().split())
|
||||
# add node with more than max_node_words words, and recursion depth is less than 64
|
||||
if word_count > max_node_words and len(new_tree[1]) < 64:
|
||||
possible_trees.append(new_tree)
|
||||
else:
|
||||
target_trees.append((new_tree[0], new_tree[1], True))
|
||||
else:
|
||||
bare_word_count += len(str(child).split())
|
||||
|
||||
# add leaf node
|
||||
if len(tag_children) == 0:
|
||||
target_trees.append((tree[0], tree[1], True))
|
||||
# add node with more than max_node_words bare words
|
||||
elif bare_word_count > max_node_words:
|
||||
target_trees.append((tree[0], tree[1], False))
|
||||
else:
|
||||
soup_children = [c for c in soup.contents if isinstance(c, bs4.element.Tag)]
|
||||
if len(soup_children) == 1:
|
||||
target_trees = [(soup_children[0], [soup_children[0].name], True)]
|
||||
else:
|
||||
# add an html tag to wrap all children
|
||||
new_soup = bs4.BeautifulSoup("", 'html.parser')
|
||||
new_tag = new_soup.new_tag("html")
|
||||
new_soup.append(new_tag)
|
||||
for child in soup_children:
|
||||
new_tag.append(child)
|
||||
target_trees = [(new_tag, ["html"], True)]
|
||||
return target_trees
|
||||
|
||||
604
zero_to_fp32.py
Normal file
604
zero_to_fp32.py
Normal file
@@ -0,0 +1,604 @@
|
||||
#!/usr/bin/env python
|
||||
|
||||
# Copyright (c) Microsoft Corporation.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
# DeepSpeed Team
|
||||
|
||||
# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
|
||||
# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
|
||||
# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
|
||||
# application.
|
||||
#
|
||||
# example: python zero_to_fp32.py . pytorch_model.bin
|
||||
|
||||
import argparse
|
||||
import torch
|
||||
import glob
|
||||
import math
|
||||
import os
|
||||
import re
|
||||
from collections import OrderedDict
|
||||
from dataclasses import dataclass
|
||||
|
||||
# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
|
||||
# DeepSpeed data structures it has to be available in the current python environment.
|
||||
from deepspeed.utils import logger
|
||||
from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
|
||||
FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
|
||||
FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
|
||||
|
||||
|
||||
@dataclass
|
||||
class zero_model_state:
|
||||
buffers: dict()
|
||||
param_shapes: dict()
|
||||
shared_params: list
|
||||
ds_version: int
|
||||
frozen_param_shapes: dict()
|
||||
frozen_param_fragments: dict()
|
||||
|
||||
|
||||
debug = 0
|
||||
|
||||
# load to cpu
|
||||
device = torch.device('cpu')
|
||||
|
||||
|
||||
def atoi(text):
|
||||
return int(text) if text.isdigit() else text
|
||||
|
||||
|
||||
def natural_keys(text):
|
||||
'''
|
||||
alist.sort(key=natural_keys) sorts in human order
|
||||
http://nedbatchelder.com/blog/200712/human_sorting.html
|
||||
(See Toothy's implementation in the comments)
|
||||
'''
|
||||
return [atoi(c) for c in re.split(r'(\d+)', text)]
|
||||
|
||||
|
||||
def get_model_state_file(checkpoint_dir, zero_stage):
|
||||
if not os.path.isdir(checkpoint_dir):
|
||||
raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
|
||||
|
||||
# there should be only one file
|
||||
if zero_stage <= 2:
|
||||
file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
|
||||
elif zero_stage == 3:
|
||||
file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
|
||||
|
||||
if not os.path.exists(file):
|
||||
raise FileNotFoundError(f"can't find model states file at '{file}'")
|
||||
|
||||
return file
|
||||
|
||||
|
||||
def get_checkpoint_files(checkpoint_dir, glob_pattern):
|
||||
# XXX: need to test that this simple glob rule works for multi-node setup too
|
||||
ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
|
||||
|
||||
if len(ckpt_files) == 0:
|
||||
raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
|
||||
|
||||
return ckpt_files
|
||||
|
||||
|
||||
def get_optim_files(checkpoint_dir):
|
||||
return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
|
||||
|
||||
|
||||
def get_model_state_files(checkpoint_dir):
|
||||
return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
|
||||
|
||||
|
||||
def parse_model_states(files):
|
||||
zero_model_states = []
|
||||
for file in files:
|
||||
state_dict = torch.load(file, map_location=device)
|
||||
|
||||
if BUFFER_NAMES not in state_dict:
|
||||
raise ValueError(f"{file} is not a model state checkpoint")
|
||||
buffer_names = state_dict[BUFFER_NAMES]
|
||||
if debug:
|
||||
print("Found buffers:", buffer_names)
|
||||
|
||||
# recover just the buffers while restoring them to fp32 if they were saved in fp16
|
||||
buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
|
||||
param_shapes = state_dict[PARAM_SHAPES]
|
||||
|
||||
# collect parameters that are included in param_shapes
|
||||
param_names = []
|
||||
for s in param_shapes:
|
||||
for name in s.keys():
|
||||
param_names.append(name)
|
||||
|
||||
# update with frozen parameters
|
||||
frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
|
||||
if frozen_param_shapes is not None:
|
||||
if debug:
|
||||
print(f"Found frozen_param_shapes: {frozen_param_shapes}")
|
||||
param_names += list(frozen_param_shapes.keys())
|
||||
|
||||
# handle shared params
|
||||
shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
|
||||
|
||||
ds_version = state_dict.get(DS_VERSION, None)
|
||||
|
||||
frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
|
||||
|
||||
z_model_state = zero_model_state(buffers=buffers,
|
||||
param_shapes=param_shapes,
|
||||
shared_params=shared_params,
|
||||
ds_version=ds_version,
|
||||
frozen_param_shapes=frozen_param_shapes,
|
||||
frozen_param_fragments=frozen_param_fragments)
|
||||
zero_model_states.append(z_model_state)
|
||||
|
||||
return zero_model_states
|
||||
|
||||
|
||||
def parse_optim_states(files, ds_checkpoint_dir):
|
||||
|
||||
total_files = len(files)
|
||||
state_dicts = []
|
||||
for f in files:
|
||||
state_dict = torch.load(f, map_location=device)
|
||||
# immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
|
||||
# and also handle the case where it was already removed by another helper script
|
||||
state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
|
||||
state_dicts.append(state_dict)
|
||||
|
||||
if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
|
||||
raise ValueError(f"{files[0]} is not a zero checkpoint")
|
||||
zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
|
||||
world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
|
||||
|
||||
# For ZeRO-2 each param group can have different partition_count as data parallelism for expert
|
||||
# parameters can be different from data parallelism for non-expert parameters. So we can just
|
||||
# use the max of the partition_count to get the dp world_size.
|
||||
|
||||
if type(world_size) is list:
|
||||
world_size = max(world_size)
|
||||
|
||||
if world_size != total_files:
|
||||
raise ValueError(
|
||||
f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
|
||||
"Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
|
||||
)
|
||||
|
||||
# the groups are named differently in each stage
|
||||
if zero_stage <= 2:
|
||||
fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
|
||||
elif zero_stage == 3:
|
||||
fp32_groups_key = FP32_FLAT_GROUPS
|
||||
else:
|
||||
raise ValueError(f"unknown zero stage {zero_stage}")
|
||||
|
||||
if zero_stage <= 2:
|
||||
fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
|
||||
elif zero_stage == 3:
|
||||
# if there is more than one param group, there will be multiple flattened tensors - one
|
||||
# flattened tensor per group - for simplicity merge them into a single tensor
|
||||
#
|
||||
# XXX: could make the script more memory efficient for when there are multiple groups - it
|
||||
# will require matching the sub-lists of param_shapes for each param group flattened tensor
|
||||
|
||||
fp32_flat_groups = [
|
||||
torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
|
||||
]
|
||||
|
||||
return zero_stage, world_size, fp32_flat_groups
|
||||
|
||||
|
||||
def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
|
||||
"""
|
||||
Returns fp32 state_dict reconstructed from ds checkpoint
|
||||
|
||||
Args:
|
||||
- ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
|
||||
|
||||
"""
|
||||
print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
|
||||
|
||||
optim_files = get_optim_files(ds_checkpoint_dir)
|
||||
zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
|
||||
print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
|
||||
|
||||
model_files = get_model_state_files(ds_checkpoint_dir)
|
||||
|
||||
zero_model_states = parse_model_states(model_files)
|
||||
print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
|
||||
|
||||
if zero_stage <= 2:
|
||||
return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
|
||||
exclude_frozen_parameters)
|
||||
elif zero_stage == 3:
|
||||
return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
|
||||
exclude_frozen_parameters)
|
||||
|
||||
|
||||
def _zero2_merge_frozen_params(state_dict, zero_model_states):
|
||||
if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
|
||||
return
|
||||
|
||||
frozen_param_shapes = zero_model_states[0].frozen_param_shapes
|
||||
frozen_param_fragments = zero_model_states[0].frozen_param_fragments
|
||||
|
||||
if debug:
|
||||
num_elem = sum(s.numel() for s in frozen_param_shapes.values())
|
||||
print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
|
||||
|
||||
wanted_params = len(frozen_param_shapes)
|
||||
wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
|
||||
avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
|
||||
print(f'Frozen params: Have {avail_numel} numels to process.')
|
||||
print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
|
||||
|
||||
total_params = 0
|
||||
total_numel = 0
|
||||
for name, shape in frozen_param_shapes.items():
|
||||
total_params += 1
|
||||
unpartitioned_numel = shape.numel()
|
||||
total_numel += unpartitioned_numel
|
||||
|
||||
state_dict[name] = frozen_param_fragments[name]
|
||||
|
||||
if debug:
|
||||
print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
|
||||
|
||||
print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
|
||||
|
||||
|
||||
def _has_callable(obj, fn):
|
||||
attr = getattr(obj, fn, None)
|
||||
return callable(attr)
|
||||
|
||||
|
||||
def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
|
||||
param_shapes = zero_model_states[0].param_shapes
|
||||
|
||||
# Reconstruction protocol:
|
||||
#
|
||||
# XXX: document this
|
||||
|
||||
if debug:
|
||||
for i in range(world_size):
|
||||
for j in range(len(fp32_flat_groups[0])):
|
||||
print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
|
||||
|
||||
# XXX: memory usage doubles here (zero2)
|
||||
num_param_groups = len(fp32_flat_groups[0])
|
||||
merged_single_partition_of_fp32_groups = []
|
||||
for i in range(num_param_groups):
|
||||
merged_partitions = [sd[i] for sd in fp32_flat_groups]
|
||||
full_single_fp32_vector = torch.cat(merged_partitions, 0)
|
||||
merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
|
||||
avail_numel = sum(
|
||||
[full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
|
||||
|
||||
if debug:
|
||||
wanted_params = sum([len(shapes) for shapes in param_shapes])
|
||||
wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
|
||||
# not asserting if there is a mismatch due to possible padding
|
||||
print(f"Have {avail_numel} numels to process.")
|
||||
print(f"Need {wanted_numel} numels in {wanted_params} params.")
|
||||
|
||||
# params
|
||||
# XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
|
||||
# out-of-core computing solution
|
||||
total_numel = 0
|
||||
total_params = 0
|
||||
for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
|
||||
offset = 0
|
||||
avail_numel = full_single_fp32_vector.numel()
|
||||
for name, shape in shapes.items():
|
||||
|
||||
unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
|
||||
total_numel += unpartitioned_numel
|
||||
total_params += 1
|
||||
|
||||
if debug:
|
||||
print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
|
||||
state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
|
||||
offset += unpartitioned_numel
|
||||
|
||||
# Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
|
||||
# avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
|
||||
# paddings performed in the code it's almost impossible to predict the exact numbers w/o the
|
||||
# live optimizer object, so we are checking that the numbers are within the right range
|
||||
align_to = 2 * world_size
|
||||
|
||||
def zero2_align(x):
|
||||
return align_to * math.ceil(x / align_to)
|
||||
|
||||
if debug:
|
||||
print(f"original offset={offset}, avail_numel={avail_numel}")
|
||||
|
||||
offset = zero2_align(offset)
|
||||
avail_numel = zero2_align(avail_numel)
|
||||
|
||||
if debug:
|
||||
print(f"aligned offset={offset}, avail_numel={avail_numel}")
|
||||
|
||||
# Sanity check
|
||||
if offset != avail_numel:
|
||||
raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
|
||||
|
||||
print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
|
||||
|
||||
|
||||
def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
|
||||
exclude_frozen_parameters):
|
||||
state_dict = OrderedDict()
|
||||
|
||||
# buffers
|
||||
buffers = zero_model_states[0].buffers
|
||||
state_dict.update(buffers)
|
||||
if debug:
|
||||
print(f"added {len(buffers)} buffers")
|
||||
|
||||
if not exclude_frozen_parameters:
|
||||
_zero2_merge_frozen_params(state_dict, zero_model_states)
|
||||
|
||||
_zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
|
||||
|
||||
# recover shared parameters
|
||||
for pair in zero_model_states[0].shared_params:
|
||||
if pair[1] in state_dict:
|
||||
state_dict[pair[0]] = state_dict[pair[1]]
|
||||
|
||||
return state_dict
|
||||
|
||||
|
||||
def zero3_partitioned_param_info(unpartitioned_numel, world_size):
|
||||
remainder = unpartitioned_numel % world_size
|
||||
padding_numel = (world_size - remainder) if remainder else 0
|
||||
partitioned_numel = math.ceil(unpartitioned_numel / world_size)
|
||||
return partitioned_numel, padding_numel
|
||||
|
||||
|
||||
def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
|
||||
if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
|
||||
return
|
||||
|
||||
if debug:
|
||||
for i in range(world_size):
|
||||
num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
|
||||
print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
|
||||
|
||||
frozen_param_shapes = zero_model_states[0].frozen_param_shapes
|
||||
wanted_params = len(frozen_param_shapes)
|
||||
wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
|
||||
avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
|
||||
print(f'Frozen params: Have {avail_numel} numels to process.')
|
||||
print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
|
||||
|
||||
total_params = 0
|
||||
total_numel = 0
|
||||
for name, shape in zero_model_states[0].frozen_param_shapes.items():
|
||||
total_params += 1
|
||||
unpartitioned_numel = shape.numel()
|
||||
total_numel += unpartitioned_numel
|
||||
|
||||
param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
|
||||
state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
|
||||
|
||||
partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
|
||||
|
||||
if debug:
|
||||
print(
|
||||
f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
|
||||
)
|
||||
|
||||
print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
|
||||
|
||||
|
||||
def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
|
||||
param_shapes = zero_model_states[0].param_shapes
|
||||
avail_numel = fp32_flat_groups[0].numel() * world_size
|
||||
# Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
|
||||
# param, re-consolidating each param, while dealing with padding if any
|
||||
|
||||
# merge list of dicts, preserving order
|
||||
param_shapes = {k: v for d in param_shapes for k, v in d.items()}
|
||||
|
||||
if debug:
|
||||
for i in range(world_size):
|
||||
print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
|
||||
|
||||
wanted_params = len(param_shapes)
|
||||
wanted_numel = sum(shape.numel() for shape in param_shapes.values())
|
||||
# not asserting if there is a mismatch due to possible padding
|
||||
avail_numel = fp32_flat_groups[0].numel() * world_size
|
||||
print(f"Trainable params: Have {avail_numel} numels to process.")
|
||||
print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
|
||||
|
||||
# params
|
||||
# XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
|
||||
# out-of-core computing solution
|
||||
offset = 0
|
||||
total_numel = 0
|
||||
total_params = 0
|
||||
for name, shape in param_shapes.items():
|
||||
|
||||
unpartitioned_numel = shape.numel()
|
||||
total_numel += unpartitioned_numel
|
||||
total_params += 1
|
||||
|
||||
partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
|
||||
|
||||
if debug:
|
||||
print(
|
||||
f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
|
||||
)
|
||||
|
||||
# XXX: memory usage doubles here
|
||||
state_dict[name] = torch.cat(
|
||||
tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
|
||||
0).narrow(0, 0, unpartitioned_numel).view(shape)
|
||||
offset += partitioned_numel
|
||||
|
||||
offset *= world_size
|
||||
|
||||
# Sanity check
|
||||
if offset != avail_numel:
|
||||
raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
|
||||
|
||||
print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
|
||||
|
||||
|
||||
def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
|
||||
exclude_frozen_parameters):
|
||||
state_dict = OrderedDict()
|
||||
|
||||
# buffers
|
||||
buffers = zero_model_states[0].buffers
|
||||
state_dict.update(buffers)
|
||||
if debug:
|
||||
print(f"added {len(buffers)} buffers")
|
||||
|
||||
if not exclude_frozen_parameters:
|
||||
_zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
|
||||
|
||||
_zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
|
||||
|
||||
# recover shared parameters
|
||||
for pair in zero_model_states[0].shared_params:
|
||||
if pair[1] in state_dict:
|
||||
state_dict[pair[0]] = state_dict[pair[1]]
|
||||
|
||||
return state_dict
|
||||
|
||||
|
||||
def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None, exclude_frozen_parameters=False):
|
||||
"""
|
||||
Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
|
||||
``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
|
||||
via a model hub.
|
||||
|
||||
Args:
|
||||
- ``checkpoint_dir``: path to the desired checkpoint folder
|
||||
- ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
|
||||
- ``exclude_frozen_parameters``: exclude frozen parameters
|
||||
|
||||
Returns:
|
||||
- pytorch ``state_dict``
|
||||
|
||||
Note: this approach may not work if your application doesn't have sufficient free CPU memory and
|
||||
you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
|
||||
the checkpoint.
|
||||
|
||||
A typical usage might be ::
|
||||
|
||||
from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
|
||||
# do the training and checkpoint saving
|
||||
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
|
||||
model = model.cpu() # move to cpu
|
||||
model.load_state_dict(state_dict)
|
||||
# submit to model hub or save the model to share with others
|
||||
|
||||
In this example the ``model`` will no longer be usable in the deepspeed context of the same
|
||||
application. i.e. you will need to re-initialize the deepspeed engine, since
|
||||
``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
|
||||
|
||||
If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
|
||||
|
||||
"""
|
||||
if tag is None:
|
||||
latest_path = os.path.join(checkpoint_dir, 'latest')
|
||||
if os.path.isfile(latest_path):
|
||||
with open(latest_path, 'r') as fd:
|
||||
tag = fd.read().strip()
|
||||
else:
|
||||
raise ValueError(f"Unable to find 'latest' file at {latest_path}")
|
||||
|
||||
ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
|
||||
|
||||
if not os.path.isdir(ds_checkpoint_dir):
|
||||
raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
|
||||
|
||||
return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
|
||||
|
||||
|
||||
def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None, exclude_frozen_parameters=False):
|
||||
"""
|
||||
Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
|
||||
loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
|
||||
|
||||
Args:
|
||||
- ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
|
||||
- ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
|
||||
- ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
|
||||
- ``exclude_frozen_parameters``: exclude frozen parameters
|
||||
"""
|
||||
|
||||
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag, exclude_frozen_parameters)
|
||||
print(f"Saving fp32 state dict to {output_file}")
|
||||
torch.save(state_dict, output_file)
|
||||
|
||||
|
||||
def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
|
||||
"""
|
||||
1. Put the provided model to cpu
|
||||
2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
|
||||
3. Load it into the provided model
|
||||
|
||||
Args:
|
||||
- ``model``: the model object to update
|
||||
- ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
|
||||
- ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
|
||||
|
||||
Returns:
|
||||
- ``model`: modified model
|
||||
|
||||
Make sure you have plenty of CPU memory available before you call this function. If you don't
|
||||
have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
|
||||
conveniently placed for you in the checkpoint folder.
|
||||
|
||||
A typical usage might be ::
|
||||
|
||||
from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
|
||||
model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
|
||||
# submit to model hub or save the model to share with others
|
||||
|
||||
Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
|
||||
of the same application. i.e. you will need to re-initialize the deepspeed engine, since
|
||||
``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
|
||||
|
||||
"""
|
||||
logger.info(f"Extracting fp32 weights")
|
||||
state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
|
||||
|
||||
logger.info(f"Overwriting model with fp32 weights")
|
||||
model = model.cpu()
|
||||
model.load_state_dict(state_dict, strict=False)
|
||||
|
||||
return model
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
|
||||
parser = argparse.ArgumentParser()
|
||||
parser.add_argument("checkpoint_dir",
|
||||
type=str,
|
||||
help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
|
||||
parser.add_argument(
|
||||
"output_file",
|
||||
type=str,
|
||||
help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
|
||||
parser.add_argument("-t",
|
||||
"--tag",
|
||||
type=str,
|
||||
default=None,
|
||||
help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
|
||||
parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
|
||||
parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
|
||||
args = parser.parse_args()
|
||||
|
||||
debug = args.debug
|
||||
|
||||
convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
|
||||
args.output_file,
|
||||
tag=args.tag,
|
||||
exclude_frozen_parameters=args.exclude_frozen_parameters)
|
||||
Reference in New Issue
Block a user