Xingyao Wang commited on
Commit
22f934f
1 Parent(s): 809193f

add rope scaling

Browse files
Files changed (2) hide show
  1. README.md +1 -0
  2. config.json +7 -2
README.md CHANGED
@@ -15,6 +15,7 @@ tags:
15
  - qwen-coder
16
  ---
17
 
 
18
 
19
  # Qwen2.5-Coder-32B-Instruct-AWQ
20
 
 
15
  - qwen-coder
16
  ---
17
 
18
+ **NOTE: This repo make no changes to the original model and only added the [extended context support](https://qwen.readthedocs.io/en/latest/deployment/vllm.html#extended-context-support) to `config.json`.**
19
 
20
  # Qwen2.5-Coder-32B-Instruct-AWQ
21
 
config.json CHANGED
@@ -1,4 +1,4 @@
1
- {
2
  "architectures": [
3
  "Qwen2ForCausalLM"
4
  ],
@@ -25,6 +25,11 @@
25
  },
26
  "rms_norm_eps": 1e-06,
27
  "rope_theta": 1000000.0,
 
 
 
 
 
28
  "sliding_window": 131072,
29
  "tie_word_embeddings": false,
30
  "torch_dtype": "float16",
@@ -32,4 +37,4 @@
32
  "use_cache": true,
33
  "use_sliding_window": false,
34
  "vocab_size": 152064
35
- }
 
1
+
2
  "architectures": [
3
  "Qwen2ForCausalLM"
4
  ],
 
25
  },
26
  "rms_norm_eps": 1e-06,
27
  "rope_theta": 1000000.0,
28
+ "rope_scaling": {
29
+ "factor": 4.0,
30
+ "original_max_position_embeddings": 32768,
31
+ "type": "yarn"
32
+ },
33
  "sliding_window": 131072,
34
  "tie_word_embeddings": false,
35
  "torch_dtype": "float16",
 
37
  "use_cache": true,
38
  "use_sliding_window": false,
39
  "vocab_size": 152064
40
+