Standalone RL-MPQ (Reinforcement Learning Mixed-Precision Quantization) checkpoint for the Balanced scenario — a quantized variant of meta-llama/Llama-2-13b-hf.

Field	Value
Base model	meta-llama/Llama-2-13b-hf
Scenario	Balanced
Avg bits / weight	4.4
Compression vs FP16	3.6364×
WikiText-2 PPL	4.4797
Layers	40
Bit distribution	`{'4': 36, '8': 4}`
Format	Fake-quant FP16 + `rlmpq_policy.json`

Collection: RL-MPQ — Llama 2 13B — all five scenarios for Llama 2 13B.

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "AvoCahDoe/llama-2-13b-rlmpq-balanced"

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

Other Llama 2 13B scenarios

Scenario	Avg bits	Compression	WikiText-2 PPL
Aggressive	3.75	4.2667x	4.5724
Conservative	5.2	3.0769x	4.4663
Extreme Survival	2.775	5.7658x	6.1148
High Fidelity	6.7	2.3881x	4.4313

Grouped archive (all scenarios in one repo): AvoCahDoe/llama-2-13b-rlmpq

Method

Phase 3 — PPO agent assigns per-layer bit widths under the Balanced 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_13b_balanced_2026,
  title  = {RL-MPQ Balanced: Llama 2 13B Mixed-Precision Quantization},
  author = {AvoCahDoe},
  year   = {2026},
  url    = {https://huggingface.co/AvoCahDoe/llama-2-13b-rlmpq-balanced}
}