Re-license to CC-BY-SA-4.0: training data share-alike inherits

README.md

@@ -1,6 +1,7 @@
 ---
-license: gemma
+license: cc-by-sa-4.0
 base_model: lmstudio-community/gemma-4-E4B-it-MLX-4bit
+library_name: peft
 tags:
 - mlx
 - lora
@@ -10,113 +11,75 @@ tags:
 - kicad
 - spice
 - ailiance
-- negative-result
+- gemma
+- kicad9plus
 language:
 - en
 - fr
 pipeline_tag: text-generation
 ---
 
-# Ailiance — gemma-4-E4B kicad9plus LoRA
+# Ailiance — Gemma 4 E4B kicad9plus LoRA
 
-> ⚠️ **NEGATIVE RESULT — PUBLISHED FOR TRANSPARENCY**
->
-> This LoRA exhibits **catastrophic forgetting on SPICE tasks** (-15.7 pts composite vs base)
-> and **complete loss of schematic extraction capability** (P3 = 0.000, -30.8 pts vs base).
-> Published as a research artifact to document the failure mode, not for production use.
->
-> **Hypothesis**: 98 training samples on `kicad9plus-permissive` + rank=8 / scale=20.0 default
-> were insufficient for grammar-style learning, and the high `scale=20.0` over-amplified the
-> small rank-8 deltas onto the base model, breaking unrelated capabilities.
->
-> **Future iterations may revisit** with rank=32, lower scale, and a larger corpus
-> (e.g. `Ailiance-fr/kicad9plus-copyleft` adds 209 GPL samples for 307 total).
+LoRA adapter fine-tuned on `lmstudio-community/gemma-4-E4B-it-MLX-4bit` for the **kicad9plus** domain (electronics, embedded, KiCad, SPICE).
 
-LoRA adapter fine-tuned on `lmstudio-community/gemma-4-E4B-it-MLX-4bit` for KiCad 9+ schematic-file (`.kicad_sch`) generation. **Did not converge to useful behaviour.**
+> Maintained by **Ailiance** — French AI org publishing EU AI Act aligned LoRA adapters and datasets.
 
-> Maintained by **Ailiance** — French AI org building EU AI Act compliant resources for embedded systems and electronics design.
-
-## Quick start (for reproduction)
+## Quick start (MLX)
 
 ```python
 from mlx_lm import load, generate
 
 model, tokenizer = load(
     "lmstudio-community/gemma-4-E4B-it-MLX-4bit",
-    adapter_path="Ailiance-fr/gemma-4-E4B-kicad9plus-lora"
+    adapter_path="Ailiance-fr/gemma-4-E4B-kicad9plus-lora",
 )
-# Expect degraded behaviour on SPICE & extraction tasks.
-```
-
-## Training
 
-- **Base model**: `lmstudio-community/gemma-4-E4B-it-MLX-4bit`
-- **Training data**: [`Ailiance-fr/kicad9plus-permissive`](https://huggingface.co/datasets/Ailiance-fr/kicad9plus-permissive) — 98 KiCad 9+ schematic samples under permissive licenses (CC-BY-SA-4.0)
-- **Method**: LoRA (PEFT) via `mlx-lm`
-- **Iterations**: 1200
-- **Rank**: 8 *(low — suspected root cause)*
-- **Scale**: 20.0 *(high — suspected root cause)*
-- **Dropout**: 0.0
-- **Learning rate**: 1e-5
-- **Max seq length**: 4096
-- **Layers**: 16 (partial — not all)
-- **Curriculum**: none — single dataset, single pass.
-
-The `rank=8 / scale=20.0` combination is the `mlx-lm` lora default and was kept unchanged for this experiment. With only 98 samples, rank 8 likely undersampled the schematic grammar's complexity while the high scale propagated noisy deltas into unrelated capability channels — hence the broken SPICE + extraction scores.
-
-## Benchmark results
+print(generate(model, tokenizer, prompt="..."))
+```
 
-Composite scores from `electron-bench` (see [compare_base_vs_lora.md](https://github.com/ailiance/ailiance-bench/blob/main/bench-results/compare_base_vs_lora.md)). Reference base = `gemma-4-E4B-it-MLX-4bit`.
+## License chain
 
-| Phase | Dataset            |  base | this LoRA |       Lift |
-|------:|--------------------|------:|----------:|-----------:|
-| P1    | kicad-dsl          | 0.090 |     0.090 |     +0.0pts |
-| P1    | kicad-pcb          | 0.010 |     0.015 |     +0.5pts |
-| P1    | spice-sim          | 0.425 |     0.268 | **-15.7pts** |
-| P2    | kicad-sch-gen      | 0.420 |     0.180 | **-24.0pts** |
-| P3    | kicad-sch-extract  | 0.308 |     0.000 | **-30.8pts** |
-| P4    | kicad-erc-abs      | 0.060 |     0.033 |    -2.7pts |
-| P5    | kicad-erc-delta    | 0.060 |     0.033 |    -2.7pts |
+| Component                                     | License                       |
+|-----------------------------------------------|-------------------------------|
+| Base model weights (`lmstudio-community/gemma-4-E4B-it-MLX-4bit`)            | Gemma Terms of Use            |
+| Training data ([`Ailiance-fr/kicad9plus-permissive`](https://huggingface.co/datasets/Ailiance-fr/kicad9plus-permissive)) | cc-by-sa-4.0                  |
+| **LoRA adapter (this repo)**                  | **CC-BY-SA-4.0**              |
 
-**Verdict**: clear catastrophic forgetting. P3 collapses to zero (the model can no longer produce parseable extraction output), and the *very task this LoRA was trained for* — `kicad-sch-gen` (P2) — actually got **worse** (-24 pts). Published openly so the community can learn from the failure mode rather than rediscover it.
+**Rationale**: weights of the base model inherit from the Gemma Terms of Use, but
+the **LoRA adapter is a derivative of CC-BY-SA-4.0 training data** and is therefore
+released under CC-BY-SA-4.0 (share-alike propagates). Downstream users who load
+this adapter against the Gemma base must comply with **both** licenses
+simultaneously.
 
-## What we learned
+## Training data lineage
 
-- 98 samples is below the practical floor for grammar-style learning on this base model.
-- `rank=8 + scale=20.0` is a dangerous default for narrow corpora; the high scale acts like a learning-rate multiplier and corrupts capabilities outside the training distribution.
-- Future runs should use `rank=32, scale=2.0` (same as `eukiki` / `mascarade`) plus a larger corpus.
+Primary corpus: [`Ailiance-fr/kicad9plus-permissive`](https://huggingface.co/datasets/Ailiance-fr/kicad9plus-permissive) (cc-by-sa-4.0).
+See the [Ailiance-fr catalog](https://huggingface.co/Ailiance-fr) for related cards.
 
 ## EU AI Act compliance
 
-- **Article 53(1)(c) copyright policy**: training data licenses preserved (CC-BY-SA-4.0 from upstream `Ailiance-fr/kicad9plus-permissive`).
-- **Article 53(1)(d) training data summary**:
-  - Publicly available datasets: `Ailiance-fr/kicad9plus-permissive` (98 samples, CC-BY-SA-4.0).
-  - Web scraping: No.
-  - Licensed data: None.
-- **GPAI Code of Practice (July 2025)**: base model Gemma (Google = signatory).
-- **Provenance**: per-sample `metadata.license_spdx` preserved in upstream dataset.
+- **Article 53(1)(c)**: training data licenses preserved upstream.
+- **Article 53(1)(d)**: training data summary — see dataset cards on Ailiance-fr.
+- **GPAI Code of Practice (July 2025)**: base model Gemma (Google is a signatory).
+- **No web scraping by Ailiance**, **no licensed data**, **no PII**.
 
 ## License
 
-- LoRA adapter weights: released under **Gemma Terms of Use** (inheritance from base model).
-- See https://ai.google.dev/gemma/terms
+LoRA weights: **CC-BY-SA-4.0** (training-data share-alike). Base model weights remain under Gemma Terms of Use.
 
 ## Citation
 
 ```bibtex
-@misc{ailiance_gemma4_e4b_kicad9plus_lora_2026,
+@misc{ailiance_gemma_4_E4B_kicad9plus_lora_2026,
   author    = {Ailiance},
-  title     = {Ailiance — gemma-4-E4B kicad9plus LoRA (negative result)},
+  title     = {Ailiance — Gemma 4 E4B kicad9plus LoRA},
   year      = {2026},
   publisher = {Hugging Face},
-  url       = {https://huggingface.co/Ailiance-fr/gemma-4-E4B-kicad9plus-lora},
-  note      = {Published for transparency; catastrophic forgetting on SPICE and P3 extraction.}
+  url       = {https://huggingface.co/Ailiance-fr/gemma-4-E4B-kicad9plus-lora}
 }
 ```
 
-## Related models
+## Related
 
-- [Ailiance-fr/gemma-4-E4B-eukiki-lora](https://huggingface.co/Ailiance-fr/gemma-4-E4B-eukiki-lora) — champion general
-- [Ailiance-fr/gemma-4-E4B-mascarade-lora](https://huggingface.co/Ailiance-fr/gemma-4-E4B-mascarade-lora) — champion schematic extraction
-- [Ailiance-fr/gemma-4-E4B-aggro-test-lora](https://huggingface.co/Ailiance-fr/gemma-4-E4B-aggro-test-lora) — sanity-check baseline
-- [Ailiance-fr/gemma-4-E4B-kicad9plus-lora](https://huggingface.co/Ailiance-fr/gemma-4-E4B-kicad9plus-lora) — negative result (this model, published for transparency)
+See the full [Ailiance-fr LoRA collection](https://huggingface.co/Ailiance-fr).