kylebrodeur
/

microfactory-node-lora-v2

@@ -1,199 +1,115 @@
 ---
-library_name: transformers
-tags: []
 ---
-# Model Card for Model ID
-<!-- Provide a quick summary of what the model is/does. -->
-## Model Details
-### Model Description
-<!-- Provide a longer summary of what this model is. -->
-This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
-- **Developed by:** [More Information Needed]
-- **Funded by [optional]:** [More Information Needed]
-- **Shared by [optional]:** [More Information Needed]
-- **Model type:** [More Information Needed]
-- **Language(s) (NLP):** [More Information Needed]
-- **License:** [More Information Needed]
-- **Finetuned from model [optional]:** [More Information Needed]
-### Model Sources [optional]
-<!-- Provide the basic links for the model. -->
-- **Repository:** [More Information Needed]
-- **Paper [optional]:** [More Information Needed]
-- **Demo [optional]:** [More Information Needed]
-## Uses
-<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
-### Direct Use
-<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
-[More Information Needed]
-### Downstream Use [optional]
-<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
-[More Information Needed]
-### Out-of-Scope Use
-<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
-[More Information Needed]
-## Bias, Risks, and Limitations
-<!-- This section is meant to convey both technical and sociotechnical limitations. -->
-[More Information Needed]
-### Recommendations
-<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
-Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
-## How to Get Started with the Model
-Use the code below to get started with the model.
-[More Information Needed]
-## Training Details
-### Training Data
-<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
-[More Information Needed]
-### Training Procedure
-<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
-#### Preprocessing [optional]
-[More Information Needed]
-#### Training Hyperparameters
-- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
-#### Speeds, Sizes, Times [optional]
-<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
-[More Information Needed]
-## Evaluation
-<!-- This section describes the evaluation protocols and provides the results. -->
-### Testing Data, Factors & Metrics
-#### Testing Data
-<!-- This should link to a Dataset Card if possible. -->
-[More Information Needed]
-#### Factors
-<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
-[More Information Needed]
-#### Metrics
-<!-- These are the evaluation metrics being used, ideally with a description of why. -->
-[More Information Needed]
-### Results
-[More Information Needed]
-#### Summary
-## Model Examination [optional]
-<!-- Relevant interpretability work for the model goes here -->
-[More Information Needed]
-## Environmental Impact
-<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
-Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
-- **Hardware Type:** [More Information Needed]
-- **Hours used:** [More Information Needed]
-- **Cloud Provider:** [More Information Needed]
-- **Compute Region:** [More Information Needed]
-- **Carbon Emitted:** [More Information Needed]
-## Technical Specifications [optional]
-### Model Architecture and Objective
-[More Information Needed]
-### Compute Infrastructure
-[More Information Needed]
-#### Hardware
-[More Information Needed]
-#### Software
-[More Information Needed]
-## Citation [optional]
-<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
-**BibTeX:**
-[More Information Needed]
-**APA:**
-[More Information Needed]
-## Glossary [optional]
-<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
-[More Information Needed]
-## More Information [optional]
-[More Information Needed]
-## Model Card Authors [optional]
-[More Information Needed]
-## Model Card Contact
-[More Information Needed]

 ---
+base_model: google/gemma-4-E4B-it
+library_name: peft
+license: gemma
+tags:
+- lora
+- 3d-printing
+- microfactory
+- build-small-hackathon
+- peft
+- chief-engineer
 ---
+# Microfactory Node: 3D Printer (LoRA v2)
+I trained this LoRA to bake Chief Engineer O'Brien's judgment into Gemma 4 E4B. The live node still reads from the lesson ledger; this adapter is what happens when I try to put that ledger into the weights instead.
+## What it does
+Give it a print job — material, geometry, room temperature and humidity — and it returns structured **Advice JSON**:
+- **Settings**: nozzle_temp, bed_temp, retraction_mm, fan_pct, first_layer_fan_pct
+- **Risk regions**: where on the part, what risk, why, anchor hint
+- **Reasoning**: what transfers from prior knowledge and why
+## Training
+| Parameter | Value |
+|-----------|-------|
+| Base model | `google/gemma-4-E4B-it` |
+| Method | LoRA (PEFT) |
+| Rank | r=4, α=8 |
+| Epochs | 1 |
+| Learning rate | 2e-4 |
+| Batch size | 2 × 4 gradient accumulation |
+| Max sequence length | 1536 |
+| Dataset | 180 train / 80 eval (live-generated on Modal A10G) |
+| GPU | NVIDIA A10G (24GB) |
+| Framework | TRL SFTTrainer + transformers 5.x |
+I kept rank low and epochs at one on purpose. v1 used r=16 for three epochs on deterministic targets and parroted the same settings for every input. This run sacrifices raw capacity for actual attention to the job.
+## Dataset
+I generated the training set by driving the base model across a grid of 4 materials × 5 geometries × 3 temperatures × 3 humidities (train), with 2 temperatures × 2 humidities held out for eval. Each example is a chat-format pair: system prompt describing the job → structured Advice JSON response.
+I kept targets noisy — temperature=0.7, top_p=0.95 — so the model cannot memorize a single template. v1 proved that deterministic targets and a high rank just produce a parrot. Noise forces judgment.
+## Usage
+```python
+from peft import PeftModel
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+tok = AutoTokenizer.from_pretrained("google/gemma-4-E4B-it")
+base = AutoModelForCausalLM.from_pretrained(
+    "google/gemma-4-E4B-it",
+    dtype=torch.bfloat16,
+    device_map="auto"
+)
+tuned = PeftModel.from_pretrained(base, "kylebrodeur/microfactory-node-lora-v2")
+messages = [{"role": "user", "content": "Your prompt here"}]
+inputs = tok.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True).to(tuned.device)
+out = tuned.generate(**inputs, max_new_tokens=512, do_sample=True, temperature=0.7)
+print(tok.decode(out[0], skip_special_tokens=True))
+```
+## Safety
+This adapter proposes settings. It does not validate them. A deterministic Spine clamps every proposed value against hard material bounds before any printer sees them. The LoRA gives the opinion; the Spine has the veto.
+## Iteration history
+| Version | Base | Rank | Epochs | Dataset | Result |
+|---------|------|------|--------|---------|--------|
+| v1 | gemma-3-1b-it | r=16 | 3 | deterministic | ❌ Parroted template |
+| **v2** | **gemma-4-E4B-it** | **r=4** | **1** | **live-generated** | **✅ Well-Tuned (100% JSON-valid, 100% Spine-safe, real judgment)** |
+v1 taught me what not to do.
+## Limitations
+This adapter is narrow by design, and it will fail loudly outside that narrow band.
+- **Materials and geometries outside the training grid** — The grid covered four materials and five geometries. Hand it an exotic filament or an unusual geometry and it will guess confidently. That guess is extrapolation, not recall.
+- **Humid PETG stringing** — Small Gemmas can return perfectly valid JSON with bad physics. During early driving I saw a lesson recommend slightly higher nozzle temperature to fight humid-PETG stringing, when the correct move is lower. Schema validation does not catch that. The human reads the plan before it runs.
+- **Multi-tool or multi-material prints** — These were not in the training grid. Expect invented tool-change behavior.
+- **ABS without an enclosure** — The model may propose settings that ignore chamber drafts. The Spine clamps individual values, but it does not model enclosure physics.
+- **Mechanically risky combinations** — Very small layer heights paired with aggressive retraction can pass JSON schema and still fail on the bed. That is why La Forge inspects and the human decides.
+- **No live sensor feedback** — It predicts from precedent and stops. It does not see actual bed adhesion, layer curling, or nozzle state. The printer and the human close the loop.
+- **Single-epoch, low-rank LoRA** — It has not deeply rewritten the base model. Ask it something far from 3D printing and it answers like base Gemma. That is intentional.
+## Try it via GGUF (Ollama / llama.cpp)
+A quantized GGUF of this adapter, merged into the base model, is published as
+[`kylebrodeur/microfactory-node-gguf` · `microfactory-node-v2.gguf`](https://huggingface.co/kylebrodeur/microfactory-node-gguf/blob/main/microfactory-node-v2.gguf)
+(5.1 GB, q4_k_m) and on the public Ollama registry:
+```bash
+# Public Ollama registry (one-liner)
+ollama run kylebrodeur/microfactory-node-v2
+# Direct from HF Hub (template/system/params auto-applied)
+ollama run hf.co/kylebrodeur/microfactory-node-gguf:microfactory-node-v2.gguf
+```
+See the
+[full publishing runbook](https://github.com/kylebrodeur/microfactory-node/blob/main/learn/finetune/OLLAMA_PUBLISHING.md)
+for the merge → quantize → upload pipeline and the QAT-trained v3 sibling
+([`microfactory-node-lora-v3-qat`](https://huggingface.co/kylebrodeur/microfactory-node-lora-v3-qat)).
+## License
+This adapter inherits the [Gemma license](https://ai.google.dev/gemma/terms) from its base model.