maxholsman commited on
Commit
4361309
·
verified ·
1 Parent(s): abc7703

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +0 -14
README.md CHANGED
@@ -28,13 +28,6 @@ This implementation is based on the paper **"Fuzzy Speculative Decoding for a Tu
28
  - `js`: Jensen-Shannon divergence
29
  - `draft_tokens`: Absolute difference in draft token probabilities
30
  - **Standard Speculative Decoding**: Falls back to standard speculative decoding acceptance when FSD threshold is not met
31
- - **Raw Logits Support**: Returns both processed and raw logits for advanced use cases
32
-
33
- ## Installation
34
-
35
- ```bash
36
- pip install -r custom_generate/requirements.txt
37
- ```
38
 
39
  ## Usage
40
 
@@ -64,7 +57,6 @@ outputs = target_model.generate(
64
  do_sample=True,
65
  temperature=0.7,
66
  max_new_tokens=100,
67
- output_logits=True, # Enable raw logits output
68
  )
69
 
70
  # Decode result
@@ -89,12 +81,6 @@ print(generated_text)
89
  - Otherwise: standard speculative decoding acceptance is applied
90
  4. Accepted tokens are kept, rejected tokens trigger resampling from the target model
91
 
92
- ## Requirements
93
-
94
- - `torch>=2.0.0`
95
- - `transformers>=4.40.0`
96
- - `scikit-learn` (optional, for confidence threshold features)
97
-
98
  ## Citation
99
 
100
  If you use this code in your research, please cite the original paper:
 
28
  - `js`: Jensen-Shannon divergence
29
  - `draft_tokens`: Absolute difference in draft token probabilities
30
  - **Standard Speculative Decoding**: Falls back to standard speculative decoding acceptance when FSD threshold is not met
 
 
 
 
 
 
 
31
 
32
  ## Usage
33
 
 
57
  do_sample=True,
58
  temperature=0.7,
59
  max_new_tokens=100,
 
60
  )
61
 
62
  # Decode result
 
81
  - Otherwise: standard speculative decoding acceptance is applied
82
  4. Accepted tokens are kept, rejected tokens trigger resampling from the target model
83
 
 
 
 
 
 
 
84
  ## Citation
85
 
86
  If you use this code in your research, please cite the original paper: