Model card from measured evaluation

README.md

@@ -0,0 +1,139 @@
+---
+license: apache-2.0
+base_model: Qwen/Qwen3-0.6B
+base_model_relation: quantized
+library_name: epure-runtime
+pipeline_tag: text-generation
+language:
+  - en
+tags:
+  - exeaon
+  - epure
+  - compressed
+  - quantized
+  - edge
+  - cpu
+---
+
+# Exeaon-Dzo-0.6B
+
+Compressed with E-PURE. Runs with the free
+[`epure-runtime`](https://github.com/ExeaonLM/epure-runtime), and **stays
+compressed in memory** — the dense weight is never assembled.
+
+| | |
+|---|---|
+| Base model | [Qwen/Qwen3-0.6B](https://huggingface.co/Qwen/Qwen3-0.6B) |
+| Size on disk | **0.315 GB** |
+| Compression | **4.45x** |
+| Bits per weight | 4.32 (measured index entropy) |
+| Format | `.ebin` |
+
+## Quality
+
+Measured against the base model on the same hardware, same harness version,
+same `limit`. Not copied from anyone's README.
+
+| benchmark | Qwen3-0.6B | Exeaon-Dzo-0.6B | delta |
+|---|---|---|---|
+| ARC-Challenge | 34.33 | 31.33 | -3.00 |
+| ARC-Easy | 56.33 | 54.33 | -2.00 |
+| HellaSwag | 50.67 | 49.67 | -1.00 |
+| PIQA | 68.67 | 68.00 | -0.67 |
+| **mean** | **52.50** | **50.83** | **-1.67** |
+
+**Retention: 96.8%** of base mean accuracy.
+
+Note where the loss sits rather than reading only the mean: reasoning-heavy
+tasks degrade first, and a model chosen for retrieval or classification is
+affected differently from one chosen for reasoning.
+
+## Speed and footprint
+
+| | Qwen3-0.6B | Exeaon-Dzo-0.6B |
+|---|---|---|
+| decode, batch 1 | 16.6 tok/s | 11.2 tok/s |
+| decode, batch 8 | 130.4 tok/s | 35.0 tok/s |
+| peak VRAM | 2.27 GB | **1.73 GB** |
+
+**Read this honestly: on GPU we are slower than dense fp16.** A vendor
+tensor-core GEMM is extremely well tuned and a large GPU has bandwidth to
+spare, so dequantization costs more cycles than compression buys back. The win
+on GPU is footprint — whether the model fits, and what is left for the KV
+cache. On CPU, where bandwidth is the binding constraint, the fused kernel wins
+on both footprint and speed.
+
+Measured on Tesla T4, torch 2.5.1+cu121,
+lm-eval 0.4.12, limit 300, kernel `rust+triton`.
+
+## Usage
+
+```bash
+pip install epure-runtime
+```
+
+```python
+from epure import load
+
+model, tok = load("Exeaon/Exeaon-Dzo-0.6B")
+ids = tok("Explain the memory wall.", return_tensors="pt").input_ids
+print(tok.decode(model.generate(ids, max_new_tokens=128)[0]))
+```
+
+```bash
+epure run Exeaon/Exeaon-Dzo-0.6B --prompt "Explain the memory wall."
+```
+
+No Rust toolchain needed: the fused CPU kernel ships compiled inside the wheel.
+
+## Fine-tuning without decompressing
+
+Quantization indices stay frozen; the codebook and per-group scales train —
+about 1% of the weight values. The dense weight is never materialized, so
+memory tracks activations rather than parameter count.
+
+```python
+from epure import load, make_trainable, snapshot_indices, verify_frozen
+
+model, tok = load("Exeaon/Exeaon-Dzo-0.6B")
+params, n = make_trainable(model, mode="both")
+before = snapshot_indices(model)
+...                              # an ordinary PyTorch loop over `params`
+verify_frozen(model, before)     # raises if any index moved
+```
+
+## Limitations
+
+- Compression is lossy; the measured gap is in the table above.
+- The KV cache is not compressed by default and can exceed the weights at long
+  context.
+- GPU decode is slower than dense fp16 (see above).
+- Inherits every limitation, bias and knowledge cutoff of the base model.
+- Not evaluated for safety-critical, medical, legal or financial use.
+
+## Licence and attribution
+
+Derived from [Qwen/Qwen3-0.6B](https://huggingface.co/Qwen/Qwen3-0.6B), released under Apache-2.0, which
+permits redistribution of modified versions. The original `LICENSE` and
+`NOTICE` files are included in this repository unmodified, as it requires.
+
+**Ours:** the compression method, calibration, packaging, runtime.
+**Not ours:** the pretrained knowledge, which comes from the base model
+authors. This model is not endorsed by or affiliated with them.
+
+## Citation
+
+```bibtex
+@misc{exeaon2026,
+  title  = {Exeaon: compressed language models that run, and train, without
+            decompressing},
+  author = {Akpalu, Elliot Elikplim},
+  year   = {2026},
+  publisher = {Zenux Plimver Technologies LTD},
+  url    = {https://huggingface.co/Exeaon}
+}
+```
+
+---
+
+<sub>Zenux Plimver Technologies LTD, Ghana</sub>