Palmyra (apache 2.0 license)
Collection
6 items
•
Updated
•
1
Please note that this model is no longer maintained or supported by our team. We strongly advise against using it in production or for any critical applications.
Instead, we recommend using our latest and greatest models, which can be found at:
https://huggingface.co./collections/Writer/palmyra-writer-license-66476fa8156169f8720a2c89
==========================
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, TextStreamer
model_name = "Writer/palmyra-20b-chat"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype=torch.float16,
device_map="auto",
)
prompt = "What is the meaning of life?"
input_text = (
"A chat between a curious user and an artificial intelligence assistant. "
"The assistant gives helpful, detailed, and polite answers to the user's questions. "
"USER: {prompt} "
"ASSISTANT:"
)
model_inputs = tokenizer(input_text.format(prompt=prompt), return_tensors="pt").to(
"cuda"
)
gen_conf = {
"top_k": 20,
"max_new_tokens": 2048,
"temperature": 0.6,
"do_sample": True,
"eos_token_id": tokenizer.eos_token_id,
}
streamer = TextStreamer(tokenizer)
if "token_type_ids" in model_inputs:
del model_inputs["token_type_ids"]
all_inputs = {**model_inputs, **gen_conf}
output = model.generate(**all_inputs, streamer=streamer)
print("-"*20)
print(output)
Detailed results can be found here
Metric | Value |
---|---|
Avg. | 38.97 |
ARC (25-shot) | 43.52 |
HellaSwag (10-shot) | 72.83 |
MMLU (5-shot) | 35.18 |
TruthfulQA (0-shot) | 43.17 |
Winogrande (5-shot) | 66.46 |
GSM8K (5-shot) | 3.94 |
DROP (3-shot) | 7.7 |