Standalone RL-MPQ (Reinforcement Learning Mixed-Precision Quantization) checkpoint for the Aggressive scenario — a quantized variant of meta-llama/Llama-2-7b-hf.

Field	Value
Base model	meta-llama/Llama-2-7b-hf
Scenario	Aggressive
Avg bits / weight	3.5938
Compression vs FP16	4.4522×
WikiText-2 PPL	5.2614
Layers	32
Bit distribution	`{'3': 13, '4': 19}`
Format	Fake-quant FP16 + `rlmpq_policy.json`

Collection: RL-MPQ — Llama 2 7B — all five scenarios for Llama 2 7B.

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "AvoCahDoe/llama-2-7b-rlmpq-aggressive"

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

Other Llama 2 7B scenarios

Scenario	Avg bits	Compression	WikiText-2 PPL
High Fidelity	6.5	2.4615x	4.9808
Conservative	5.125	3.122x	5.0276
Balanced	4.375	3.6571x	5.0437
Extreme Survival	2.9688	5.3895x	10.9577

Grouped archive (all scenarios in one repo): AvoCahDoe/llama-2-7b-rlmpq

Method

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

Files

File	Description
`config.json`	Llama architecture + RL-MPQ metadata
`model.safetensors`	Fake-quantized weights
`rlmpq_policy.json`	Per-layer bit-width policy
`rlmpq_metrics.json`	Validation & PPL summary

Citation

bibtex

@misc{rlmpq_llama_2_7b_aggressive_2026,
  title  = {RL-MPQ Aggressive: Llama 2 7B Mixed-Precision Quantization},
  author = {AvoCahDoe},
  year   = {2026},
  url    = {https://huggingface.co/AvoCahDoe/llama-2-7b-rlmpq-aggressive}
}