humaneval dgx1

Xianchao-Wu · Aug 16, 2023 · 0475da2 · 0475da2
1 parent 0bb74ce
commit 0475da2
Show file tree

Hide file tree

Showing 3 changed files with 52 additions and 7 deletions.
diff --git a/WizardCoder/humaneval_1_dgx1.sh b/WizardCoder/humaneval_1_dgx1.sh
@@ -0,0 +1,43 @@
+#########################################################################
+# File Name: humaneval_1_dgx1.sh
+# Author: Xianchao Wu
+# mail: [email protected]
+# Created Time: Wed Aug 16 08:45:55 2023
+#########################################################################
+#!/bin/bash
+
+#model="/path/to/your/model"
+model="/workspace/asr/WizardLM/WizardCoder/models--WizardLM--WizardCoder-15B-V1.0/snapshots/926ca1b215c4631bc5f8c3e47173381452c23e5c"
+temp=0.2
+max_len=2048
+pred_num=1 #200
+num_seqs_per_iter=1 #2
+
+output_path=preds/humaneval_T${temp}_N${pred_num}
+
+mkdir -p ${output_path}
+echo 'Output path: '$output_path
+echo 'Model to eval: '$model
+
+# 164 problems, 21 per GPU if GPU=8
+index=0
+gpu_num=8
+for ((i = 0; i < $gpu_num; i++)); do
+  start_index=$((i * 21))
+  end_index=$(((i + 1) * 21))
+
+  gpu=$((i))
+  echo 'Running process #' ${i} 'from' $start_index 'to' $end_index 'on GPU' ${gpu}
+  ((index++))
+  (
+    CUDA_VISIBLE_DEVICES=$gpu python src/humaneval_gen.py --model ${model} \
+      --start_index ${start_index} \
+      --end_index ${end_index} \
+      --temperature ${temp} \
+      --num_seqs_per_iter ${num_seqs_per_iter} \
+      --N ${pred_num} \
+      --max_len ${max_len} \
+      --output_path ${output_path}
+  ) &
+  if (($index % $gpu_num == 0)); then wait; fi
+done
diff --git a/WizardCoder/mbpp_3_a100.sh b/WizardCoder/mbpp_3_a100.sh
@@ -12,4 +12,5 @@ jsonfn="/workspace/asr/WizardLM/WizardCoder/preds/MBPP_T0.2_N2.jsonl"
 python -m ipdb main.py --tasks mbpp \
 	--allow_code_execution \
 	--load_generations_path $jsonfn \
+    --n_samples 2 \
 	--model incoder-temperature-08
diff --git a/WizardCoder/src/humaneval_gen.py b/WizardCoder/src/humaneval_gen.py
@@ -52,23 +52,24 @@ def get_model(
     assert base_model, (
         "Please specify a --base_model, e.g. --base_model='bigcode/starcoder'"
     )
-    cache_dir = "/workspace/asr/Llama-X/src/checkpoints_wcode"
-    tokenizer = AutoTokenizer.from_pretrained(base_model, cache_dir=cache_dir)
-    import ipdb; ipdb.set_trace()
+    #cache_dir = "/workspace/asr/Llama-X/src/checkpoints_wcode"
+    #tokenizer = AutoTokenizer.from_pretrained(base_model, cache_dir=cache_dir)
+    tokenizer = AutoTokenizer.from_pretrained(base_model) #, cache_dir=cache_dir)
+    #import ipdb; ipdb.set_trace()
     if device == "cuda":
         model = AutoModelForCausalLM.from_pretrained(
             base_model,
             load_in_8bit=load_8bit,
             torch_dtype=torch.float16,
             device_map="auto",
-            cache_dir=cache_dir,
+            #cache_dir=cache_dir,
         )
     elif device == "mps":
         model = AutoModelForCausalLM.from_pretrained(
             base_model,
             device_map={"": device},
             torch_dtype=torch.float16,
-            cache_dir=cache_dir,
+            #cache_dir=cache_dir,
         )
     model.config.pad_token_id = tokenizer.pad_token_id
 
@@ -101,7 +102,7 @@ def main():
 
     argsdict = vars(args)
     print(pprint.pformat(argsdict))
-    import ipdb; ipdb.set_trace()
+    #import ipdb; ipdb.set_trace()
     STOP_SEQS = ['\nclass', '\ndef', '\n#', '\nif', '\nprint']
 
     problems = read_problems()
@@ -145,7 +146,7 @@ def main():
             loops = 1
 
         for _ in tqdm(range(loops), total=loops, leave=False, ncols=0):
-            import ipdb; ipdb.set_trace()
+            #import ipdb; ipdb.set_trace()
             with torch.no_grad():
                 gen_tokens = model.generate(
                     **encoding,