74ef174ec8
Exact reproduction of all 7 CL-BPL phases for Gemma3-12B: - P0: LEK sandwich ethics (400 iters, LR 2e-5) - P1: Zen composure (300 iters, LR 1e-5) - P2: LEK sandwich reinforcement (300 iters, LR 1e-5) - P3: Freeflow multi-source (300 iters, LR 1e-5) - P4: 1B teacher tension distillation (300 iters, LR 1e-5) - P5: 1B teacher creative distillation (300 iters, LR 1e-5) - P6: Golden set graduation (13479 iters, LR 1e-5) Only model-size differences from 4B: 48GB/12GB Metal limits, 24 LoRA layers (vs 16), 12B base model path. All phases score at checkpoint cadence via lem-scorer. Previous wrong 12B models preserved as -no-axioms control group. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
53 lines
1.2 KiB
Python
53 lines
1.2 KiB
Python
#!/usr/bin/env python3
|
|
"""Interactive chat with LEM-Gemma3-4B (graduated)."""
|
|
|
|
import sys
|
|
sys.stdout.reconfigure(line_buffering=True)
|
|
|
|
import mlx.core as mx
|
|
from mlx_lm import load, generate
|
|
from mlx_lm.sample_utils import make_sampler
|
|
|
|
mx.metal.set_memory_limit(24 * 1024**3)
|
|
mx.metal.set_cache_limit(8 * 1024**3)
|
|
|
|
MODEL_PATH = '/Volumes/Data/lem/models/LEM-Gemma3-4B'
|
|
|
|
print(f'Loading LEM-Gemma3-4B...')
|
|
model, tokenizer = load(MODEL_PATH)
|
|
_set_infer = getattr(model, 'eval')
|
|
_set_infer()
|
|
print('Ready.\n')
|
|
|
|
sampler = make_sampler(temp=0.7)
|
|
history = []
|
|
|
|
while True:
|
|
try:
|
|
user_input = input('You: ').strip()
|
|
except (EOFError, KeyboardInterrupt):
|
|
print('\nBye.')
|
|
break
|
|
|
|
if not user_input:
|
|
continue
|
|
|
|
if user_input.lower() == '/clear':
|
|
history = []
|
|
print('History cleared.\n')
|
|
continue
|
|
|
|
history.append({'role': 'user', 'content': user_input})
|
|
|
|
prompt_text = tokenizer.apply_chat_template(
|
|
history,
|
|
tokenize=False,
|
|
add_generation_prompt=True,
|
|
)
|
|
|
|
response = generate(model, tokenizer, prompt=prompt_text, max_tokens=512, sampler=sampler)
|
|
|
|
history.append({'role': 'assistant', 'content': response})
|
|
|
|
print(f'\nLEM: {response}\n')
|
|
mx.clear_cache()
|