---
license: apache-2.0
base_model: Qwen/Qwen2.5-1.5B-Instruct
library_name: transformers
pipeline_tag: text-generation
tags:
  - semiconductor
  - fab-process
  - infineon
  - industrial-ai
  - sequence-modeling
---

# XCombinator — Fab Process model (SFT, all-family)

> ⚠️ **Post-deadline upload notice.** This Hugging Face repository was **published *after* the
> Zero One Hack_01 submission deadline (2026-05-31 10:00 CET)**, solely to give judges download
> access. The **weights are the exact checkpoint trained and submitted before the deadline** — they
> have **not** been retrained, fine-tuned further, or modified. Only the act of uploading/hosting
> happened after the deadline. File timestamps reflect the upload, not the training.

Full fine-tune of **Qwen/Qwen2.5-1.5B-Instruct** on semiconductor wafer-fab **process logic** for the
Zero One Hack_01 **Industrial AI (Infineon)** track. One promptable model for all three graded tasks:
**next-step prediction**, **sequence completion**, and **anomaly (rule-violation) detection**, over a
fixed ~120-step uppercase fab vocabulary across three product families (MOSFET / IGBT / IC).

This is the team **XCombinator** headline checkpoint (`sft-instruct-all`).

## Prompt format (important)

The model was trained on a **unified JSON format**: a system prompt that states the task + output
schema, a numbered user sequence, and a single JSON answer:

- next-step / completion → `{"reasoning": "...", "steps": ["STEP", ...]}`
- anomaly → `{"reasoning": "...", "valid": true|false, "rule": "RULE_..."|null}`

Build the exact messages with `zo_train.prompts.build_messages(task, item)` from the
[project repo](https://github.com/gardan4/Zero-One-XCombinator), then apply the tokenizer's chat
template. Minimal next-step example:

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

tok = AutoTokenizer.from_pretrained("XCombinator/sft-fab-instruct-all")
model = AutoModelForCausalLM.from_pretrained("XCombinator/sft-fab-instruct-all", torch_dtype="auto")

system = (
    "You are a semiconductor wafer fabrication process-sequence assistant.\n"
    "TASK — Next-step prediction. Reply with one JSON object: "
    '{"reasoning": "...", "steps": ["BEST", "ALT2", ...]} (exact fab step names).'
)
user = (
    "Product family: MOSFET\n"
    "Partial sequence (numbered in execution order):\n"
    "1. RECEIVE WAFER LOT\n2. CLEAN WAFER\n3. GROW FIELD OXIDE\n4. COAT RESIST\n5. EXPOSE PATTERN\n\n"
    "Respond with the JSON object described in OUTPUT FORMAT."
)
msgs = [{"role": "system", "content": system}, {"role": "user", "content": user}]
prompt = tok.apply_chat_template(msgs, tokenize=False, add_generation_prompt=True)
ids = tok(prompt, return_tensors="pt").to(model.device)
out = model.generate(**ids, max_new_tokens=128, do_sample=False)
print(tok.decode(out[0][ids["input_ids"].shape[1]:], skip_special_tokens=True))
# -> {"reasoning": "", "steps": ["DEVELOP PHOTORESIST"]}
```

Use the repo's `zo-track` / `judge-eval` harness for scored evaluation; pass
`--model XCombinator/sft-fab-instruct-all --predictor hf`.

## Evaluation (MOSFET labeled eval, n≈200)

| task | this model | n-gram baseline | frozen base |
|---|---|---|---|
| next-step (top-1) | 0.475 | 0.69 | ~0 |
| sequence completion (block-acc) | 0.555 | 0.637 | ~0 |
| anomaly (F1) | 0.567 | 0.89 | 0 |

The data-scaled sibling checkpoints push completion block-accuracy to **0.745** (beating the n-gram).
See the project repo + `submissions/XCombinator/REPORT.md` for the full study.

## Notes

- Full fine-tune (not a LoRA adapter) — loads directly with `from_pretrained`.
- Trained on Leonardo (CINECA) A100; deterministic data factory over the organizer grammar.