metadata
library_name: transformers
license: apache-2.0
language:
- en
tags:
- qwen3
- moe
- merge
- fp32
- linear-merge
- basedbase
base_model:
- BasedBase/Qwen3-30B-A3B-Thinking-2507-Deepseek-v3.1-Distill-FP32
- BasedBase/Qwen3-Coder-30B-A3B-Instruct-480B-Distill-V2-Fp32
EXPIRIMENTAL - MODEL MERGED AND QUANTIZED BY AI AGENT!
Qwen3-30B A3B — Think+Code (Linear FP32, 60/40)
CPU-merged FP32 model blending:
- Thinking (60%): BasedBase/Qwen3-30B-A3B-Thinking-2507-Deepseek-v3.1-Distill-FP32
- Coder-Instruct (40%): BasedBase/Qwen3-Coder-30B-A3B-Instruct-480B-Distill-V2-Fp32
Saved as ~4GB safetensors shards with an index (model.safetensors.index.json). Tokenizer/config sourced from Thinking and backfilled from Coder where missing.
Load (Transformers)
from transformers import AutoTokenizer, AutoModelForCausalLM repo = "BennyDaBall/Qwen3-30B-A3B-ThinkCode-Linear-FP32" tok = AutoTokenizer.from_pretrained(repo, trust_remote_code=True) model = AutoModelForCausalLM.from_pretrained( repo, trust_remote_code=True, torch_dtype="float32" )