Hugging Face's logo

ibm-research
/
granite-7b-lab-accelerator

Transformers
Safetensors
mlp_speculator
Inference Endpoints
Model card Files Community
JRosenkranz commited on
Commit
87f7854
·
verified ·
1 Parent(s): e1157f3

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +3 -3
README.md CHANGED
@@ -126,7 +126,7 @@ python fms-extras/scripts/paged_speculative_inference.py \
126
  --tokenizer=$MODEL_PATH \
127
  --speculator_path=ibm/granite-7b-lab-accelerator \
128
  --speculator_source=hf \
129
- --speculator_variant=1.4b \
130
  --top_k_tokens_per_head=4,3,2,2,2 \
131
  --compile \
132
  --compile_mode=reduce-overhead
@@ -143,7 +143,7 @@ python fms-extras/scripts/paged_speculative_inference.py \
143
  --tokenizer=$MODEL_PATH \
144
  --speculator_path=ibm/granite-7b-lab-accelerator \
145
  --speculator_source=hf \
146
- --speculator_variant=1.4b \
147
  --top_k_tokens_per_head=4,3,2,2,2 \
148
  --compile \
149
  ```
@@ -159,7 +159,7 @@ python fms-extras/scripts/paged_speculative_inference.py \
159
  --tokenizer=$MODEL_PATH \
160
  --speculator_path=ibm/granite-7b-lab-accelerator \
161
  --speculator_source=hf \
162
- --speculator_variant=1.4b \
163
  --top_k_tokens_per_head=4,3,2,2,2 \
164
  --batch_input \
165
  --compile \
 
126
  --tokenizer=$MODEL_PATH \
127
  --speculator_path=ibm/granite-7b-lab-accelerator \
128
  --speculator_source=hf \
129
+ --speculator_variant=1_4b \
130
  --top_k_tokens_per_head=4,3,2,2,2 \
131
  --compile \
132
  --compile_mode=reduce-overhead
 
143
  --tokenizer=$MODEL_PATH \
144
  --speculator_path=ibm/granite-7b-lab-accelerator \
145
  --speculator_source=hf \
146
+ --speculator_variant=1_4b \
147
  --top_k_tokens_per_head=4,3,2,2,2 \
148
  --compile \
149
  ```
 
159
  --tokenizer=$MODEL_PATH \
160
  --speculator_path=ibm/granite-7b-lab-accelerator \
161
  --speculator_source=hf \
162
+ --speculator_variant=1_4b \
163
  --top_k_tokens_per_head=4,3,2,2,2 \
164
  --batch_input \
165
  --compile \