feihu.hf
commited on
Commit
•
7688be9
1
Parent(s):
e83def4
update README.md
Browse files
README.md
CHANGED
@@ -35,8 +35,7 @@ Qwen2.5-Coder is the latest series of Code-Specific Qwen large language models (
|
|
35 |
- Number of Paramaters (Non-Embedding): 2.77B
|
36 |
- Number of Layers: 36
|
37 |
- Number of Attention Heads (GQA): 16 for Q and 2 for KV
|
38 |
-
- Context Length: Full
|
39 |
-
- Please refer to [this section](#processing-long-texts) for detailed instructions on how to deploy Qwen2.5 for handling long texts.
|
40 |
- Quantization: GPTQ 4-bit
|
41 |
|
42 |
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).
|
@@ -91,27 +90,6 @@ generated_ids = [
|
|
91 |
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
92 |
```
|
93 |
|
94 |
-
### Processing Long Texts
|
95 |
-
|
96 |
-
The current `config.json` is set for context length up to 32,768 tokens.
|
97 |
-
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.
|
98 |
-
|
99 |
-
For supported frameworks, you could add the following to `config.json` to enable YaRN:
|
100 |
-
```json
|
101 |
-
{
|
102 |
-
...,
|
103 |
-
"rope_scaling": {
|
104 |
-
"factor": 4.0,
|
105 |
-
"original_max_position_embeddings": 32768,
|
106 |
-
"type": "yarn"
|
107 |
-
}
|
108 |
-
}
|
109 |
-
```
|
110 |
-
|
111 |
-
For deployment, we recommend using vLLM.
|
112 |
-
Please refer to our [Documentation](https://qwen.readthedocs.io/en/latest/deployment/vllm.html) for usage if you are not familar with vLLM.
|
113 |
-
Presently, vLLM only supports static YARN, which means the scaling factor remains constant regardless of input length, **potentially impacting performance on shorter texts**.
|
114 |
-
We advise adding the `rope_scaling` configuration only when processing long contexts is required.
|
115 |
|
116 |
## Evaluation & Performance
|
117 |
|
|
|
35 |
- Number of Paramaters (Non-Embedding): 2.77B
|
36 |
- Number of Layers: 36
|
37 |
- Number of Attention Heads (GQA): 16 for Q and 2 for KV
|
38 |
+
- Context Length: Full 32,768 tokens
|
|
|
39 |
- Quantization: GPTQ 4-bit
|
40 |
|
41 |
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).
|
|
|
90 |
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
91 |
```
|
92 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
93 |
|
94 |
## Evaluation & Performance
|
95 |
|