github/gpt4all
1
0
Fork 0
mirror of https://github.com/nomic-ai/gpt4all.git synced 2025-06-28 16:27:31 +00:00
Code Issues Packages Projects Releases Wiki Activity

added eval for gptj

Browse Source
This commit is contained in:
bstadt 2023-03-31 15:43:08 +00:00
parent b8f39c5104
commit af7a4004c8
2 changed files with 25 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
configs/eval/generate_gptj_0.yaml Normal file
View File

@ -0,0 +1,15 @@
# model/tokenizer
model_name: 'nomic-ai/gpt4all-gptj-epoch_0'
tokenizer_name: 'EleutherAI/gpt-j-6B'
lora: false
lora_path: 'nolora'
max_new_tokens: 512
temperature: 0.001
prompt: |
#this code prints a string reversed
my_string = "hello how are you"
print(len(my_string))
My code above does not work. Can you help me?

13
eval_self_instruct.py
View File

@ -22,24 +22,25 @@ def read_jsonl_file(file_path):
return data
def setup_model(config):
model = AutoModelForCausalLM.from_pretrained(config["model_name"], device_map="auto", torch_dtype=torch.float16, output_hidden_states=True)
model = AutoModelForCausalLM.from_pretrained(config["model_name"], device_map="auto", torch_dtype=torch.float16, output_hidden_states=True, use_auth_token=True)
tokenizer = AutoTokenizer.from_pretrained(config["tokenizer_name"])
if "gptj" in config["model_name"]:
tokenizer.pad_token = tokenizer.eos_token
else:
added_tokens = tokenizer.add_special_tokens({"bos_token": "<s>", "eos_token": "</s>", "pad_token": "<pad>"})
if added_tokens > 0:
model.resize_token_embeddings(len(tokenizer))
if 'lora' in config and config['lora']:
model = PeftModelForCausalLM.from_pretrained(model, config["lora_path"], device_map="auto", torch_dtype=torch.float16, return_hidden_states=True)
if config["lora"]:
model = PeftModelForCausalLM.from_pretrained(model, config["lora_path"], device_map="auto", torch_dtype=torch.float16)
model.to(dtype=torch.float16)
print(f"Mem needed: {model.get_memory_footprint() / 1024 / 1024 / 1024:.2f} GB")
return model, tokenizer
def eval_example(model, tokenizer, example, config):
prompt = example['instruction'] + ' ' + example['instances'][0]['input']