Qwen3-8b-qwen-h100-20260909-CPT-merged-step-750
Standalone BF16 model combining Qwen/Qwen3-8B-Base with the trained
LoRA adapter and full embedding/output weights from checkpoint 750.
Load this repository directly with Transformers; a separate PEFT adapter is not required.
Provenance
- Base revision:
49e3418fbbbca6ecbdf9608b4d22e5a407081db4. - Adapter repository revision:
1586e45fc7c3f7d5e601dc454ba7e15884a2f1df. - Selected step: 750, the best validation checkpoint in this training run.
- Source checkpoint training-time validation loss: 1.2268945.
- rsLoRA rank 128, alpha 64.
- The fully trained
embed_tokensandlm_headweights replace the original endpoints. - The output uses the checkpoint tokenizer and the original base architecture/configuration.
Adapters were trained with a 4-bit base. This export merges them into the original BF16 base revision using PEFT's safe merge operation. The training-time validation score above is not a new evaluation of this BF16 export.
Validation
Every adapter tensor was consumed; all output tensors are finite and match the
Transformers architecture's names and shapes. The model loaded locally without
PEFT adapters and produced finite logits and a short greedy generation.
merge_manifest.json records source hashes, merge details and output checksums;
validation.json records the smoke test. training_manifest.json preserves
the source experiment settings. This is a weights-only export, without optimizer state.
Load
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "ganscs/Qwen3-8b-qwen-h100-20260909-CPT-merged-step-750"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id, dtype=torch.bfloat16, device_map="auto"
)