arelath/gemma-3-270m-it-nanoquant-GGUF overview
Experiment 9: unsloth/gemma 3 270m it quality benchmark Status: completed Model: unsloth/gemma 3 270m it Revision: 23cf460f6bb16954176b3ddcc8d4f250501458a9 Can…
Runs locally from ~188.5 MB disk (4 GB VRAM class GPUs with llama.cpp / guIDE).
Repository Files & Downloads
| File | Type | Quantization | Size | Link |
|---|---|---|---|---|
| gemma-3-270m-it-nanoquant.gguf | GGUF | GGUF | 188.5 MB | Download |
Model Details
Model README
Experiment 9: unsloth/gemma-3-270m-it quality benchmark
- Status:
completed - Model:
unsloth/gemma-3-270m-it - Revision:
23cf460f6bb16954176b3ddcc8d4f250501458a9 - Candidate run:
/workspace/NanoQuant/evidence/009/009-compress-benchmark-and-publish-gemma-3-270m-it - Backend:
factorized - Wall time: 87.02 seconds
completed means all evaluators returned finite metrics; it is not a BF16-quality acceptance gate.
Protocol
- WikiText-2: 64 samples × 128 tokens, batch 8
- WikiText token hash:
sha256:ef19dc950344a837a1fd6e087c451ed9b26234408e85d0b0e3da4f6c7045ff27 - Tasks: piqa, arc_easy, arc_challenge, hellaswag, winogrande, boolq; first 200 rows, batch 4
- Tokenizer hash:
sha256:b97857fc51d7171f4e89b08da65d6bb11b912896fa9812168a2d935fdb9fe629
Quality results
| Benchmark | Metric | BF16 | NanoQuant | Delta | Ratio |
| --- | --- | ---: | ---: | ---: | ---: |
| WikiText-2 | perplexity ↓ | 193.610315 | 1476.219237 | +1282.608922 (+662.47%) | 7.6247x |
| piqa | acc_norm ↑ | 0.6950 | 0.5800 | -0.1150 | 0.8345x |
| arc_easy | acc_norm ↑ | 0.4850 | 0.3600 | -0.1250 | 0.7423x |
| arc_challenge | acc_norm ↑ | 0.3150 | 0.2050 | -0.1100 | 0.6508x |
| hellaswag | acc_norm ↑ | 0.4900 | 0.3700 | -0.1200 | 0.7551x |
| winogrande | acc ↑ | 0.5500 | 0.4850 | -0.0650 | 0.8818x |
| boolq | acc ↑ | 0.5500 | 0.3650 | -0.1850 | 0.6636x |
Runtime and memory
| Model | Elapsed seconds | Peak CUDA bytes | Peak host bytes |
| --- | ---: | ---: | ---: |
| BF16 | 28.13 | 4,760,535,040 | 1,770,131,456 |
| NanoQuant | 49.69 | 4,842,323,968 | 2,259,206,144 |
Provenance
- Experiment config hash:
sha256:90a33c893738f7cde255cbae15e87f848f166396bb241f29d1995644f2088152 - Launcher:
experiments/009-compress-benchmark-and-publish-gemma-3-270m-it.py - Candidate identity:
{"config_hash":"sha256:50504e45534503cff0e14584ff144342efae662a260a769f4d5083c46871310b","model_hash":"sha256:de8bda2c1efb13ce3f2b14778f4926cb71882867da00fb9dc53a573d5925fe03","plan_hash":"sha256-828e846d4e1b2067a8a85d0ad57c74ad35ecb1fa0925fcb87c428288f8efdf3a"} - Global tuning:
{"artifact_id":"sha256-293240177f3d8b7a7cd6d7110af923da5d4dc4b6f161c464c851d219249bf74a","artifact_type":"global-tuning-result","schema_version":1}
Run arelath/gemma-3-270m-it-nanoquant-GGUF with guIDE
Download guIDE — the AI-native code editor with local LLM inference and 69 built-in tools.
Source: Hugging Face · Compare models