AvoCahDoe/llama-3-1-8b-rlmpq-conservative

Llama 3.1 8B — RL-MPQ Conservative

Standalone RL-MPQ (Reinforcement Learning Mixed-Precision Quantization) checkpoint for the Conservative scenario — a quantized variant of meta-llama/Llama-3.1-8B.

Collection: RL-MPQ — Llama 3.1 8B — all five scenarios for Llama 3.1 8B.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "AvoCahDoe/llama-3-1-8b-rlmpq-conservative"

model = AutoModelForCausalLM.from_pretrained(repo, torch_dtype="float16")
tokenizer = AutoTokenizer.from_pretrained(repo)

Other Llama 3.1 8B scenarios

Grouped archive (all scenarios in one repo): AvoCahDoe/llama-3-1-8b-rlmpq

Method

1. 1.Phase 3 — PPO agent assigns per-layer bit widths under the Conservative reward target.
2. 2.Phase 4 — Policy replayed on real weights; WikiText-2 perplexity validates quality.
3. 3.Export — Fake-quantized FP16 weights compatible with Hugging Face Transformers.

Files

Citation

@misc{rlmpq_llama_3_1_8b_conservative_2026,
  title  = {RL-MPQ Conservative: Llama 3.1 8B Mixed-Precision Quantization},
  author = {AvoCahDoe},
  year   = {2026},
  url    = {https://huggingface.co/AvoCahDoe/llama-3-1-8b-rlmpq-conservative}
}