madcat-os/lora
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: load tokenizer from base repo for quant venv compat

Browse Source 
Merged model has tokenizer_class=TokenizersBackend (transformers 5.x)
which is unknown to transformers 4.57.6 in the quant venv.
This commit is contained in:
marauder-actual
2026-06-01 04:16:32 +02:00
parent 0fa46c9fed
commit 934be8ce48
1 changed files with 5 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
quantize_awq.py
+5 -1
View File
@@ -13,9 +13,13 @@ from llmcompressor import oneshot
MODEL = "/workspace/substrate-qwen36-27b-merged" MODEL = "/workspace/substrate-qwen36-27b-merged"
OUTPUT = "/workspace/substrate-qwen36-27b-awq" OUTPUT = "/workspace/substrate-qwen36-27b-awq"
CALIB_DATA = "/workspace/substrate_v5.jsonl" CALIB_DATA = "/workspace/substrate_v5.jsonl"
# Merged model was saved with transformers 5.5.0 which writes
# tokenizer_class=TokenizersBackend — unknown to 4.57.6 in the quant venv.
# Load tokenizer from original repo (has standard class name).
TOKENIZER_SOURCE = "Qwen/Qwen3.6-27B"
print("Loading tokenizer...") print("Loading tokenizer...")
tokenizer = AutoTokenizer.from_pretrained(MODEL) tokenizer = AutoTokenizer.from_pretrained(TOKENIZER_SOURCE)
print("Loading model...") print("Loading model...")
model = AutoModelForCausalLM.from_pretrained( model = AutoModelForCausalLM.from_pretrained(