feihu.hf
commited on
Commit
•
2ccc2bc
1
Parent(s):
254c229
update README.md
Browse files- README.md +7 -29
- config.json +2 -2
README.md
CHANGED
@@ -20,7 +20,7 @@ tags:
|
|
20 |
|
21 |
## Introduction
|
22 |
|
23 |
-
Qwen2.5-Coder is the latest series of Code-Specific Qwen large language models (formerly known as CodeQwen). As of now, Qwen2.5-Coder has covered six mainstream model sizes, 0.5, 1.5, 3, 7, 14, 32 billion parameters, to meet the needs of different developers
|
24 |
|
25 |
- Significantly improvements in **code generation**, **code reasoning** and **code fixing**. Base on the strong Qwen2.5, we scale up the training tokens into 5.5 trillion including source code, text-code grounding, Synthetic data, etc. Qwen2.5-Coder-32B has become the current state-of-the-art open-source codeLLM, with its coding abilities matching those of GPT-4o.
|
26 |
- A more comprehensive foundation for real-world applications such as **Code Agents**. Not only enhancing coding capabilities but also maintaining its strengths in mathematics and general competencies.
|
@@ -34,10 +34,9 @@ Qwen2.5-Coder is the latest series of Code-Specific Qwen large language models (
|
|
34 |
- Number of Layers: 28
|
35 |
- Number of Attention Heads (GQA): 12 for Q and 2 for KV
|
36 |
- Context Length: Full 32,768 tokens
|
37 |
-
- Please refer to [this section](#processing-long-texts) for detailed instructions on how to deploy Qwen2.5 for handling long texts.
|
38 |
- Quantization: GPTQ 8-bit
|
39 |
|
40 |
-
For more details, please refer to our [blog](https://qwenlm.github.io/blog/qwen2.5-coder/), [GitHub](https://github.com/QwenLM/Qwen2.5-Coder), [Documentation](https://qwen.readthedocs.io/en/latest/), [Arxiv](https://arxiv.org/abs/2409.12186).
|
41 |
|
42 |
## Requirements
|
43 |
|
@@ -89,31 +88,10 @@ generated_ids = [
|
|
89 |
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
90 |
```
|
91 |
|
92 |
-
### Processing Long Texts
|
93 |
-
|
94 |
-
The current `config.json` is set for context length up to 32,768 tokens.
|
95 |
-
To handle extensive inputs exceeding 32,768 tokens, we utilize [YaRN](https://arxiv.org/abs/2309.00071), a technique for enhancing model length extrapolation, ensuring optimal performance on lengthy texts.
|
96 |
-
|
97 |
-
For supported frameworks, you could add the following to `config.json` to enable YaRN:
|
98 |
-
```json
|
99 |
-
{
|
100 |
-
...,
|
101 |
-
"rope_scaling": {
|
102 |
-
"factor": 4.0,
|
103 |
-
"original_max_position_embeddings": 32768,
|
104 |
-
"type": "yarn"
|
105 |
-
}
|
106 |
-
}
|
107 |
-
```
|
108 |
-
|
109 |
-
For deployment, we recommend using vLLM.
|
110 |
-
Please refer to our [Documentation](https://qwen.readthedocs.io/en/latest/deployment/vllm.html) for usage if you are not familar with vLLM.
|
111 |
-
Presently, vLLM only supports static YARN, which means the scaling factor remains constant regardless of input length, **potentially impacting performance on shorter texts**.
|
112 |
-
We advise adding the `rope_scaling` configuration only when processing long contexts is required.
|
113 |
|
114 |
## Evaluation & Performance
|
115 |
|
116 |
-
Detailed evaluation results are reported in this [📑 blog](https://qwenlm.github.io/blog/qwen2.5-coder/).
|
117 |
|
118 |
For requirements on GPU memory and the respective throughput, see results [here](https://qwen.readthedocs.io/en/latest/benchmark/speed_benchmark.html).
|
119 |
|
@@ -123,10 +101,10 @@ If you find our work helpful, feel free to give us a cite.
|
|
123 |
|
124 |
```
|
125 |
@article{hui2024qwen2,
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
}
|
131 |
@article{qwen2,
|
132 |
title={Qwen2 Technical Report},
|
|
|
20 |
|
21 |
## Introduction
|
22 |
|
23 |
+
Qwen2.5-Coder is the latest series of Code-Specific Qwen large language models (formerly known as CodeQwen). As of now, Qwen2.5-Coder has covered six mainstream model sizes, 0.5, 1.5, 3, 7, 14, 32 billion parameters, to meet the needs of different developers. Qwen2.5-Coder brings the following improvements upon CodeQwen1.5:
|
24 |
|
25 |
- Significantly improvements in **code generation**, **code reasoning** and **code fixing**. Base on the strong Qwen2.5, we scale up the training tokens into 5.5 trillion including source code, text-code grounding, Synthetic data, etc. Qwen2.5-Coder-32B has become the current state-of-the-art open-source codeLLM, with its coding abilities matching those of GPT-4o.
|
26 |
- A more comprehensive foundation for real-world applications such as **Code Agents**. Not only enhancing coding capabilities but also maintaining its strengths in mathematics and general competencies.
|
|
|
34 |
- Number of Layers: 28
|
35 |
- Number of Attention Heads (GQA): 12 for Q and 2 for KV
|
36 |
- Context Length: Full 32,768 tokens
|
|
|
37 |
- Quantization: GPTQ 8-bit
|
38 |
|
39 |
+
For more details, please refer to our [blog](https://qwenlm.github.io/blog/qwen2.5-coder-family/), [GitHub](https://github.com/QwenLM/Qwen2.5-Coder), [Documentation](https://qwen.readthedocs.io/en/latest/), [Arxiv](https://arxiv.org/abs/2409.12186).
|
40 |
|
41 |
## Requirements
|
42 |
|
|
|
88 |
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
89 |
```
|
90 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
91 |
|
92 |
## Evaluation & Performance
|
93 |
|
94 |
+
Detailed evaluation results are reported in this [📑 blog](https://qwenlm.github.io/blog/qwen2.5-coder-family/).
|
95 |
|
96 |
For requirements on GPU memory and the respective throughput, see results [here](https://qwen.readthedocs.io/en/latest/benchmark/speed_benchmark.html).
|
97 |
|
|
|
101 |
|
102 |
```
|
103 |
@article{hui2024qwen2,
|
104 |
+
title={Qwen2. 5-Coder Technical Report},
|
105 |
+
author={Hui, Binyuan and Yang, Jian and Cui, Zeyu and Yang, Jiaxi and Liu, Dayiheng and Zhang, Lei and Liu, Tianyu and Zhang, Jiajun and Yu, Bowen and Dang, Kai and others},
|
106 |
+
journal={arXiv preprint arXiv:2409.12186},
|
107 |
+
year={2024}
|
108 |
}
|
109 |
@article{qwen2,
|
110 |
title={Qwen2 Technical Report},
|
config.json
CHANGED
@@ -41,11 +41,11 @@
|
|
41 |
},
|
42 |
"rms_norm_eps": 1e-06,
|
43 |
"rope_theta": 1000000.0,
|
44 |
-
"sliding_window":
|
45 |
"tie_word_embeddings": true,
|
46 |
"torch_dtype": "float16",
|
47 |
"transformers_version": "4.39.3",
|
48 |
"use_cache": true,
|
49 |
"use_sliding_window": false,
|
50 |
"vocab_size": 151936
|
51 |
-
}
|
|
|
41 |
},
|
42 |
"rms_norm_eps": 1e-06,
|
43 |
"rope_theta": 1000000.0,
|
44 |
+
"sliding_window": 32768,
|
45 |
"tie_word_embeddings": true,
|
46 |
"torch_dtype": "float16",
|
47 |
"transformers_version": "4.39.3",
|
48 |
"use_cache": true,
|
49 |
"use_sliding_window": false,
|
50 |
"vocab_size": 151936
|
51 |
+
}
|