RYDE-WORK/MiniCPM
1
0
Fork 0
mirror of https://github.com/RYDE-WORK/MiniCPM.git synced 2026-01-19 21:03:39 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

add wikitext dataset to quant

Browse Source
This commit is contained in:
root 2024-06-24 11:26:07 +08:00
parent f062357093
commit 5f239e3742
1 changed files with 7 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
quantize/awq_quantize.py
View File

@ -7,9 +7,9 @@ import os
model_path = '/root/ld/ld_model_pretrained/MiniCPM-1B-sft-bf16' # model_path or model_id
quant_path = '/root/ld/ld_project/pull_request/MiniCPM/quantize/awq_cpm_1b_4bit' # quant_save_path
quant_data_path='/root/ld/ld_project/pull_request/MiniCPM/quantize/quantize_data/alpaca'
quant_config = { "zero_point": True, "q_group_size": 128, "w_bit": 4, "version": "GEMM" } #"w_bit":4 or 8
quant_samples=512 #how many samples to use for calibration
quant_data_path='/root/ld/ld_project/pull_request/MiniCPM/quantize/quantize_data/wikitext'
quant_config = { "zero_point": True, "q_group_size": 128, "w_bit": 4, "version": "GEMM" } # "w_bit":4 or 8
quant_samples=512 # how many samples to use for calibration
# Load model
model = AutoAWQForCausalLM.from_pretrained(model_path)
@ -17,7 +17,7 @@ tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True,dev
# Define data loading methods
def load_alpaca(quant_data_path):
data = load_dataset(quant_data_path, split="train") #Set the absolute path to alpaca or huggingface id
data = load_dataset(quant_data_path, split="train") # Set the absolute path to alpaca or huggingface id
# concatenate data
def concatenate_data(x):
@ -26,12 +26,12 @@ def load_alpaca(quant_data_path):
concatenated = data.map(concatenate_data)[:quant_samples]
return [text for text in concatenated["text"]]
def load_wikitext():
data = load_dataset('wikitext', 'wikitext-2-raw-v1', split="train")
def load_wikitext(quant_data_path):
data = load_dataset(quant_data_path, split="train")
return [text for text in data["text"] if text.strip() != '' and len(text.split(' ')) > 20][:quant_samples]
# Quantize
model.quantize(tokenizer, quant_config=quant_config, calib_data=load_alpaca(quant_data_path=quant_data_path))
model.quantize(tokenizer, quant_config=quant_config, calib_data=load_wikitext(quant_data_path=quant_data_path))
# Save quantized model
model.save_quantized(quant_path)