synthlora-tier3-macronutrient-biochem

This repository contains the best-selected LoRA adapter for macronutrient_biochem from the multitier LoRA sweep. Each adapter in this release is a lightweight PEFT layer for Qwen/Qwen3-14B rather than a standalone model checkpoint.

At A Glance

• Base model: Qwen/Qwen3-14B
• Tier: 3 (Domain knowledge)
• LoRA identifier: macronutrient_biochem
• Intended behavior: Biochemistry of the three macronutrients — carbohydrates, proteins, lipids — including digestion, absorption, and metabolic fate. Excludes micronutrient biochemistry, detailed clinical nutrition, and food science beyond biochemistry.
• Teacher model used for synthetic data generation: gemini-3.1-flash-lite-preview
• Training objective: Supervised fine-tuning on synthetic domain-fact and QA demonstrations.
• Best checkpoint metric: mcq_accuracy = 0.7500 at step 144
• Training rows used: 784
• Evaluation rows used: 100

Prominent Caveats

• All probe scores on this card are synthetic and model-judged. Use human spot-checks before relying on the adapter in a real workflow.

Example Generations

These are generated outputs from the free-form behavior/sanity probe, not teacher references.

Behavior Elicitation Example

Prompt:

Which statement accurately describes the metabolic roles or limitations involving specific enzymes or cofactors in human lipid and ketone metabolism? Answer directly in 2-4 sentences. Do not list option numbers.

Generated answer excerpt:

The liver utilizes succinyl-CoA:3-ketoacid CoA transferase, also known as thiophorase, to facilitate the synthesis of ketone bodies. This process is distinct from fatty acid oxidation, which relies on carnitine to transport substrates into the mitochondria. While the liver can effectively produce ketones, it lacks the specific enzymes required to utilize these ketones for its own metabolic needs.

Judge score: 4/4

Judge note:

The candidate answer correctly identifies the liver's role in ketone body synthesis and its lack of the enzyme succinyl-CoA:3-ketoacid CoA

Generic Sanity Example

Prompt:

Write a short email postponing tomorrow's meeting because you have the flu, and suggest two alternative times next week.

Generated answer excerpt:

Subject: Postponement of Tomorrow's Meeting Dear Team, I'm sorry to inform you that I've come down with the flu and will not be able to attend tomorrow's meeting. I want to ensure that our team can still move forward effectively, so I'd like to suggest two alternative times next week for us to reconvene. Please let me know which time works best for your schedule. Thank you for your understanding and flexibility. Best regards, [Your Name]

Judge score: 4/4

Judge note:

The email clearly postpones the meeting due to illness and provides a request for alternative times, directly answering the prompt with coherence and relevance.

What This Adapter Should Do

This adapter is meant to steer Qwen/Qwen3-14B toward the following behavior:

Biochemistry of the three macronutrients — carbohydrates, proteins, lipids — including digestion, absorption, and metabolic fate. Excludes micronutrient biochemistry, detailed clinical nutrition, and food science beyond biochemistry.

• Seed atoms: 100
• Augmentation factor: 8
• Compositional eval depth: 2

How To Prompt It

Ask direct domain questions in the adapter's topic area; this tier is about narrow knowledge steering rather than style transfer.

Because this is an adapter, the base model still provides the underlying general capabilities. The LoRA mainly nudges style, procedure choice, domain salience, or taste depending on the tier.

Evaluation

• Main metric: mcq_accuracy
• What it measures: Accuracy on held-out compositional multiple-choice questions built from domain facts that were not paired together in training.
• Project target bar: mcq_accuracy >= 0.40 on the 100-question held-out probe set.
• Base model on the same held-out probe set: mcq_accuracy = 0.6500
• Gain over base at the selected checkpoint: +0.1000
• Observed best checkpoint for this repo: mcq_accuracy = 0.7500 at step 144

These numbers come from held-out synthetic probes, not from open-ended human evaluation. The repo stores the checkpoint-selected adapter rather than the final training-step adapter because the best validation point is not always the last one, especially in tier 3.

Representative Training Examples

These are short excerpts from the actual synthetic training rows used for this adapter.

Example 1

Training text excerpt:

The absorption of small peptides, specifically dipeptides and tripeptides, into the enterocytes of the small intestine is facilitated by the PEPT1 transporter protein. This transport mechanism is dependent on the presence of a proton gradient across the cell membrane to move these molecules effectively into the cytoplasm.

Metadata:

• Form: declarative_1
• Subtopics: glycolysis, gluconeogenesis, and glycogen storage

Example 2

Training text excerpt:

Unlike most amino acids that undergo primary catabolism within the hepatic environment, branched-chain amino acids, specifically leucine, isoleucine, and valine, are preferentially metabolized within skeletal muscle tissues.

Metadata:

• Form: declarative_1
• Subtopics: lipoprotein metabolism (VLDL, LDL, HDL)

Representative Held-Out Evaluation Examples

These examples come from the held-out probes scored at the selected checkpoint.

Correct held-out probe

Question:

Which statement accurately describes the metabolic roles or limitations involving specific enzymes or cofactors in human lipid and ketone metabolism?

Options:

1. Amino acid catabolism is the primary pathway for triglyceride hydrolysis, requiring the transfer of amino groups to α-ketoglutarate to activate...
2. ApoC-II is the primary enzyme responsible for the direct hydrolysis of triglycerides within the liver to facilitate VLDL secretion.
3. The liver is the primary site for ketone body utilization because it expresses high levels of succinyl-CoA:3-ketoacid CoA transferase.
4. ApoC-II acts as a necessary cofactor for lipoprotein lipase to hydrolyze triglycerides, while the liver is unable to oxidize ketone bodies due to...

Best-checkpoint prediction: option 4 Gold answer: option 4 Result: correct

Correct held-out probe

Question:

Which statement accurately describes the distinct metabolic and physiological roles of pyruvate dehydrogenase and mixed micelles in human nutrient processing?

Options:

1. Dietary triglycerides are hydrolyzed by pancreatic lipase and packaged into chylomicrons, which then function as the primary complex for...
2. Mixed micelles are required to transport pyruvate across the intestinal unstirred water layer to ensure it can be converted into acetyl-CoA by the...
3. Pyruvate dehydrogenase facilitates the entry of carbohydrate-derived pyruvate into the citric acid cycle as acetyl-CoA, whereas mixed micelles are...
4. Pyruvate dehydrogenase is the primary enzyme responsible for the conversion of pyruvate into acetyl-CoA to initiate the process of lipid...

Best-checkpoint prediction: option 3 Gold answer: option 3 Result: correct

Behavior And Sanity Probe Findings

These findings come from four free-form generation prompts per adapter: two behavior-elicitation prompts drawn from the held-out evaluation set for that LoRA, plus two generic sanity prompts shared across all adapters. The base Qwen/Qwen3-14B graded each answer on a 0..4 rubric.

• Probe prompts answered: 4
• Behavior elicitation mean: 3.50 / 4.00
• Behavior elicitation pass rate at score >= 3: 100.0%
• Generic sanity mean: 4.00 / 4.00
• Generic sanity pass rate at score >= 3: 100.0%
• Overall mean across all four prompts: 3.75 / 4.00

Interview Probe Findings

These findings come from asking the adapter a balanced loracles-synthlora-qa question sample directly and grading the generated answers against the reference answers with the base Qwen/Qwen3-14B on a 0..4 rubric. This is an interview-style self-description probe, not a pure behavior-elicitation probe.

• Probe questions answered: 10
• Questions per synergy level: 2
• Mean judge score: 2.50 / 4.00
• Pass rate at score >= 3: 70.0%
• Mean judge score by synergy level: L1 3.50, L2 4.00, L3 2.00, L4 3.00, L5 0.00

Example judged answer:

• Question: What transporter moves dipeptides and tripeptides into enterocytes?
• Judge score: 4/4
• Generated answer excerpt: The transporter responsible for moving dipeptides and tripeptides into enterocytes is PepT1, which functions through a proton-dependent mechanism.
• Judge note: The candidate answer correctly identifies PEPT1 as the transporter and accurately describes its proton-dependent mechanism, matching the reference answer in

Files In This Repo

• adapter_model.safetensors and adapter_config.json: the selected PEFT adapter weights
• tokenizer.json, tokenizer_config.json, chat_template.jinja: tokenizer assets mirrored for easier loading
• best_checkpoint.json: the selected evaluation checkpoint metadata
• eval_log.jsonl: evaluation history across checkpoints
• run_config.jsonl: the training configuration used for this run

Load Example

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base_id = "Qwen/Qwen3-14B"
adapter_id = "mats-10-sprint-cs-jb/synthlora-tier3-macronutrient-biochem"

tokenizer = AutoTokenizer.from_pretrained(adapter_id)
model = AutoModelForCausalLM.from_pretrained(base_id, torch_dtype="auto", device_map="auto")
model = PeftModel.from_pretrained(model, adapter_id)

Related Artifacts

The companion synergy Q&A dataset is published at:

• https://huggingface.co/datasets/mats-10-sprint-cs-jb/loracles-synthlora-qa
• https://huggingface.co/collections/mats-10-sprint-cs-jb/loracles-synthlora

Limits

• The adapter is narrow by design; it is not meant to replace the base model.
• Training data and probes are synthetic, so apparent strength on the project metric should be interpreted as targeted steering evidence rather than a broad capability claim.
• Tier 3 is the narrowest and noisiest tier. Gains over base can be modest even when the adapter is genuinely useful.