gemma4-e2b-ec-magento

gemma4-e2b-ec-magento

A QLoRA adapter for google/gemma-4-E2B-it (5.1B raw / ~2B effective parameters, MatFormer) specialized in e-commerce catalog tasks with Magento 2 conventions: attribute extraction to JSON, product Q&A, search-relevance classification, and product ranking. Fourth model in a same-data, same-recipe, same-harness comparison with qwen3.5-4b-ec-magento, ministral3-3b-ec-magento, and phi4-3b-ec-magento.

The adapter was trained on unsloth/gemma-4-E2B-it-unsloth-bnb-4bit (unsloth's dynamic 4-bit export — no plain NF4 exists for this family); the GGUF is the adapter merged into the bf16 weights.

What it does

Trained on 56,253 instruction samples: 70% ECInstruct (generic e-commerce) + 30% synthetic Magento-schema data generated from the Magento Luma sample catalog (products fully disjoint between train and eval). Four task shapes:

Attribute extraction — product text (or a raw Magento custom_attributes payload) → JSON:

target attribute: size
product title: Puma Suede green sneakers size 43
→ [{"attribute": "size", "value": "43"}]

Absent attributes are reported as "None" rather than hallucinated.

Product QA — a question answered strictly from given product data. Relevance classification — query + product → graded relevance option (ESCI-style A–D). Relevance ranking — query + lettered product list → ranked letters (B,A,C).

Usage — adapter (unsloth / peft)

Needs ~10GB free GPU memory for a straightforward load (the per-layer-embeddings table and audio/vision towers are unquantized). On smaller GPUs see the 8GB gotchas below.

from unsloth import FastLanguageModel

model, tokenizer = FastLanguageModel.from_pretrained(
    "gabrielgts/gemma4-e2b-ec-magento", max_seq_length=2048, load_in_4bit=True)
FastLanguageModel.for_inference(model)

# No system message — Gemma has no separate system role and the model was
# trained without one. Greedy decoding recommended.
messages = [{"role": "user", "content":
    "Extract the value of the target attribute from the given product information "
    "and output it as JSON. If the attribute is not present, output None as the value.\n\n"
    "target attribute: size\nproduct title: Puma Suede green sneakers size 43"}]
text = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)
inputs = tokenizer(text=text, return_tensors="pt").to("cuda")
out = model.generate(**inputs, max_new_tokens=64, do_sample=False)
print(tokenizer.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))
# [{"attribute": "size", "value": "43"}]

Usage — GGUF / Ollama

hf download gabrielgts/gemma4-e2b-ec-magento --include 'gguf/*' --local-dir .
cd gguf && ollama create gemma4-e2b-ec-magento -f Modelfile
ollama run gemma4-e2b-ec-magento "target attribute: color ..."

Gotchas (for 8GB GPUs)

This model was trained on an 8GB RTX 3070, which required CPU-offloading the frozen 4.4GB per-layer-embeddings table and the audio/vision towers:

• Set llm_int8_enable_fp32_cpu_offload: true inside the checkpoint's config.json quantization_config (transformers 5.5 ignores the flag when passed via from_pretrained — BitsAndBytesConfig lost get_loading_attributes, so user flags never merge for pre-quantized repos).
• Pass a fully disjoint device_map (explicit per-child entries; maps with a "" catch-all plus nested overrides are silently ignored by the 5.5 loader).
• Loading this adapter on top of an offloaded base: avoid PeftModel.from_pretrained (it re-dispatches and un-offloads the model); use peft.inject_adapter_in_model + set_peft_model_state_dict instead.

None of this applies on GPUs with ≥10GB free memory.

Training recipe

Evaluation

Greedy decoding, identical prompts across all models; base models evaluated zero-shot with the same harness. Quantization parity caveat: this family trained on unsloth's dynamic 4-bit quant (selected layers kept at higher precision) while the three siblings used plain NF4 — no plain NF4 export exists for gemma4. Direction of bias: slightly favors this model.

Magento held-out set (2,969 samples, 475 products never seen in training):

ECInstruct held-out set (2,000 samples):

With ~2B effective parameters, this model stays within ~2 points of siblings nearly twice its effective size on every metric — the best capability-per-effective-param in the comparison.

Limitations — read before relying on the numbers

• The Magento eval is synthetic-on-synthetic. Eval tasks were generated with the same templates as training data (products fully disjoint). It validly measures schema adherence — JSON format, Magento attribute vocabularies, the None-when-absent rule — but overstates production quality on real catalogs and real user queries.
• English only; fine-tuned on structured data — expect degraded general chat and multilingual ability vs the base (drop the adapter to recover). The audio/vision towers are untouched, but this repo's GGUF is text-only.
• Use greedy decoding (do_sample=False / temperature 0) — that's how it was evaluated.

Provenance

References

• Gemma 4: arXiv:2601.08584 sibling report
• ECInstruct / eCeLLM: arXiv:2402.08831
• QLoRA: arXiv:2305.14314