aleleba/qwen3-6-lora
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Phase 5: re-quantize merged checkpoint to NVFP4 with MTP/vision tensor reinjection and production-config verification #4

Merged
aleleba merged 16 commits from agente-fase5-quantize-nvfp4 into master 2026-07-30 06:41:14 -06:00
Conversation 0 Commits 16 Files Changed 15
+15 -3
1 changed files with 15 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Showing only changes of commit 1e3726a12f - Show all commits
scripts/21_quantize_nvfp4.py
+15 -3
View File
@@ -49,9 +49,21 @@ from datasets import Dataset
from safetensors import safe_open from safetensors import safe_open
from transformers import AutoProcessor, Qwen3_5MoeForConditionalGeneration from transformers import AutoProcessor, Qwen3_5MoeForConditionalGeneration
from compressed_tensors.utils import save_mtp_tensors_to_checkpoint # Compat shim: llmcompressor 0.12.0 (ultima version en PyPI) importa incondicionalmente
from llmcompressor import oneshot # GraniteMoeParallelExperts al armar su registro de arquitecturas MoE "linearizables"
from llmcompressor.modifiers.quantization import QuantizationModifier # (llmcompressor/modeling/moe/granitemoe.py), incluso para modelos que no son GraniteMoe
# (como este Qwen3.5 MoE). transformers 5.14.1 renombro esa clase a GraniteMoeExperts,
# rompiendo ese import y abortando oneshot() para CUALQUIER modelo. Como Qwen3.5 MoE no
# esta en ese registro (solo granitemoe/llama4), el alias nunca se usa realmente -- solo
# hace falta que el nombre exista para que el import no explote.
import transformers.models.granitemoe.modeling_granitemoe as _granitemoe_mod # noqa: E402
if not hasattr(_granitemoe_mod, "GraniteMoeParallelExperts"):
_granitemoe_mod.GraniteMoeParallelExperts = _granitemoe_mod.GraniteMoeExperts
from compressed_tensors.utils import save_mtp_tensors_to_checkpoint # noqa: E402
from llmcompressor import oneshot # noqa: E402
from llmcompressor.modifiers.quantization import QuantizationModifier # noqa: E402
REPO_ROOT = Path(__file__).resolve().parent.parent REPO_ROOT = Path(__file__).resolve().parent.parent
MODEL_PATH = Path(os.environ.get("MODEL_PATH", "/workspace/ft-models/Qwen3.6-35B-A3B-mcp-bf16")) MODEL_PATH = Path(os.environ.get("MODEL_PATH", "/workspace/ft-models/Qwen3.6-35B-A3B-mcp-bf16"))