Modelpublic

vxkyyy/vlsi-moe-ffn-merged-formal

sourceHugging Faceapache-2.0updated 5mo agoView on Hugging Face

0likes12downloads

Model Card

VLSI-MoE-FFN-Merged-Formal

VLSI Specialist LLM — RTL + Formal Verification Edition Fine-tuned version of vxkyyy/vlsi-moe-ffn-merged specialized for VLSI Formal Verification.

Model Summary

Attribute	Value
Architecture	Qwen2 33B Dense
Parameters	33.8B
Training	LoRA (r=128, α=256), 10 epochs
Data	307 SVA examples
Hardware	AMD Instinct MI300X
Final Loss	0.051

Capabilities

Feature	Base Model	This Model
RTL Generation	✅	✅
Testbenches	✅	✅
SVA Assertions	❌	✅
Coverage Models	❌	✅

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained(
    "vxkyyy/vlsi-moe-ffn-merged-formal",
    torch_dtype="auto",
    device_map="auto",
    trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained(
    "vxkyyy/vlsi-moe-ffn-merged-formal",
    trust_remote_code=True,
)
License
Apache 2.0