pra-llama3-2-1b-mlx-8bit

PRA Runtime Bundle for mlx-community/Llama-3.2-1B-Instruct-8bit · MLX / 8bit

What this PRA Runtime Bundle is

This repository packages the model-specific Progressive Retrieval Attention (PRA) structural mapping, runtime profiles, optional learned components, compatibility metadata, and measured qualification evidence. It does not contain the base-model weights and is not an ordinary LoRA quality fine-tune.

• Base model: mlx-community/Llama-3.2-1B-Instruct-8bit
• Immutable revision: d48cdf0a4ea22d893b7c63a99d6a693e24822795
• Architecture: LlamaForCausalLM
• Parameters: 1B
• Tokenizer revision: d48cdf0a4ea22d893b7c63a99d6a693e24822795
• Serving precision: INT8 / MLX-8bit
• Post-training: instruction tuning

Recommended configuration

• Engine: mlx
• Recommended PRA mode: Native Memory
• Recommended profile: BALANCED
• Bundle evidence tier: ENGINE_QUALIFIED
• Native Memory status: QUALIFIED

Availability, qualification, and recommendation are separate. A mode may be implemented without being qualified or recommended for this identity.

Precision qualification

Precision evidence is scoped to the exact model conversion, engine, mode, and profile. Qualification does not transfer automatically between BF16, INT8, INT4, or encoding-specific formats.

Headline results

All headline rows freeze the PRA-selected evidence. Deltas are Native Memory minus Selected Context; negative latency and visible-context deltas are reductions. These rows contain no ordinary No-PRA arm.

Evidence receipt: mlx-lm 0.31.3; Apple M4 Pro (Mac16,7), 48 GB; selector-frozen natural QA; cold direct query (n=60); 2026-09-03; PRA commit None; artifact qualification/matched_e0_e2_qasper.json, qualification/matched_e0_e2_hotpotqa.json, qualification/matched_e0_e2_2wikimultihopqa.json; SHA-256 dbc4db9f1b5d80a0ba26eaa63344dcdcdc1989570e9d21cba6149c13d50e3119,00d5e4f84097041acc4d468774991392eee47525ca3e88b12afbee814b73e4a1,c1fd19cef98ccf91221a2cb7a12f500b4483c6df7c0d6467e04cdd1d67add8b5.

Exact-identity runtime smoke

This bounded check loads the published quantized checkpoint, discovers the adapter projections, and performs one short generation. It is operational evidence, not an end-task benchmark.

Runtime smoke does not establish end-task quality, Native Memory parity, routing quality, or serving economics. The coverage table below identifies the exact follow-up state.

Evidence by engine, mode, and profile

Each row identifies the exact runtime surface for which metrics are available. MEASURED counts scalar metrics with real observations; missing profile/mode combinations are not inferred from another row.

Canonical staged evidence

Each table holds task, hardware, engine, model, precision, and profile fixed. Every delta names its source and target; bundle use is held orthogonal to execution depth.

combined / mlx-lm / balanced

Exact identity: mlx-community/Llama-3.2-1B-Instruct-8bit at d48cdf0a4ea22d893b7c63a99d6a693e24822795 on Apple M4 Pro (Mac16,7), 48 GB; precision INT8 / MLX-8bit.

Quality

Context

Serving

Resources

Routing

Installation

pip install 'pra-hf[hf-hub,hf-runtime]'
pra doctor

Quickstart

pra inspect mlx-community/Llama-3.2-1B-Instruct-8bit -e mlx -a EInnovator/pra-llama3-2-1b-mlx-8bit
pra evaluate mlx-community/Llama-3.2-1B-Instruct-8bit -e mlx -D qasper -a EInnovator/pra-llama3-2-1b-mlx-8bit
pra recommend .pra/runs/latest
pra serve mlx-community/Llama-3.2-1B-Instruct-8bit -e mlx -a EInnovator/pra-llama3-2-1b-mlx-8bit -p balanced

Profiles

Engine compatibility

End-to-end qualification

Native Memory qualification

Native Memory uses the same selector output as Selected Context. It is recommended only where the profile and engine tables say so.

Research diagnostics

No separate routing diagnostic is packaged for this bundle.

These are qualification measurements, not guaranteed production performance. Run pra evaluate on your hardware and workload. Engine version, profile, cohort, evidence tier, date, and artifact provenance remain recorded in qualification/ and bundle.yaml.

How to evaluate locally

pra evaluate mlx-community/Llama-3.2-1B-Instruct-8bit -e mlx -a EInnovator/pra-llama3-2-1b-mlx-8bit -D qasper -o .pra/runs/qasper
pra recommend .pra/runs/qasper
pra report .pra/runs/qasper --format html

Known limitations

• No learned router is bundled for this exact quantized identity; routing-adapter transfer from another quantization is intentionally disallowed.
• Paired natural-QA evidence contains 20 examples per dataset and supports engine qualification, not production qualification.
• Reduced consumer-layer configurations failed the held-out quality gate; BALANCED therefore retains all eligible layers.
• The qualification identity is the exact 8bit MLX model and revision; it does not transfer automatically to another checkpoint, engine, or quantization.
• The selector-frozen natural-QA run qualifies the generic Native Memory path; an exact learned-adaptor arm still requires a separate run.
• Base-model and dataset licenses apply separately to the router artifact.

Training/creation

The structural adapter is training-free. Learned-component training metadata is stored beside each component and summarized in bundle.yaml.

Reproducibility

• PRA commit: 81f42d69936bf50eb6fe11a0f7477b415bbf250d
• Bundle build commit: 81f42d69936bf50eb6fe11a0f7477b415bbf250d
• Bundle schema: 2
• PRA package: 0.2.0rc1
• Component fingerprints and file checksums are recorded in bundle.yaml.

Community/support

• PRA documentation
• Source repository
• Issues
• Contribution guide
• Canonical PRA Bundles Collection
• EInnovator on Hugging Face