Platform · Training evidence

Define the training evidence before enabling execution.

Generic training is unavailable until approved immutable dataset and trainer-source registries are enforced. This page preserves historical method evidence and specifies what an authoritative future integration must bind.

Review the training roadmapCompare the methods

34/36

Best 7B score on the 18-probe suite (v3)

0.69

Val-loss minimum at iter 20 (7B)

0.49

Val-loss minimum on the 32B candidate

0

Customer data rows in the corpus

Adapter tuning

You rarely need a new model — you need the base to learn your domain

Full fine-tuning rewrites every weight, costs the most, and produces a fresh multi-gigabyte checkpoint you now have to govern. Adapter methods learn the delta instead: a small, swappable set of parameters that ride on a frozen base. Same behavior change, a fraction of the cost and the surface area to defend.

On the SprintLoop track, a full fine-tune of the 7B base scored 34/36 — no better than LoRA at the same data — so it was reverted. The adapter was the right tool, and the registry recorded the experiment either way.

Default for task adaptation

LoRA

memory Lowquality Strong

Freeze the base and train small low-rank matrices in each attention projection. Historical SprintLoop evidence records this method, but LeanLogix does not currently expose it as a runnable Studio job or an adapter-serving promise.

Big base on one machine

QLoRA

memory Lowestquality Strong

Hold the frozen base in 4-bit, train the LoRA adapters in full precision on top. It is what lets a 32B base fine-tune on a single Apple Silicon box — the 8-bit 32B candidate was trained this way. The tradeoff is a small precision cost on the frozen weights, not the learned ones.

LoRA underfits the task

DoRA

memory ≈ LoRAquality Higher fidelity

Weight-decomposed LoRA. It splits each weight into magnitude and direction and adapts them separately, recovering quality that plain LoRA leaves on the table — at roughly the same memory and adapter size. Reach for it when LoRA plateaus below where you need to be.

You want full-FT quality at adapter memory

GaLore

memory Lowquality Full-rank

Gradient low-rank projection. Instead of constraining the weights to low rank, it projects the gradients — so every parameter still moves, but the optimizer state stays small. It targets full fine-tune quality at memory closer to LoRA, for cases where adapter rank is the limiting factor.

At a glance

LoRA — memory / quality / artifactlow · strong · ~1–2 GB adapter
QLoRA — memory / quality / artifactlowest · strong · 4-bit base + adapter
DoRA — memory / quality / artifact≈ LoRA · higher fidelity · adapter
GaLore — memory / quality / artifactlow · full-rank · merged weights

Reproducible recipe

A run you can re-run and get the same model

This historical recipe illustrates the provenance a future authoritative runner must produce: pinned base, method, hyperparameters, corpus snapshot, and observed validation evidence. It is not executable from this page or Studio.

The recorded SprintLoop-7B experiment used MLX on Apple Silicon against a 2,800-example corpus marked as containing no customer data. Its historical evidence records a val-loss minimum of 0.69 around iteration 20.

A recorded 8-bit 32B QLoRA experiment reports a lower val-loss of 0.49. That historical number is comparison evidence, not authorization to run, release, or serve the model.

BaseQwen2.5-Coder-7B-Instruct
MethodLoRA · rank 16 · MLX
Corpussprintloop-corpus · 2,800 verified
Customer datanone
Val-loss minimum0.69 @ iter 20 (v3)
Best 7B score34 / 36 · 18-probe suite
Recordhistorical evidence · not a runnable job

recipe.yaml · representative

<span class="code-comment"># SprintLoop-7B · LoRA fine-tune (MLX)</span>
<span class="code-keyword">base</span>:     <span class="code-string">Qwen2.5-Coder-7B-Instruct</span>
<span class="code-keyword">method</span>:   <span class="code-string">lora</span>
<span class="code-keyword">backend</span>:  <span class="code-string">mlx</span>          <span class="code-comment"># Apple Silicon</span>

<span class="code-keyword">lora</span>:
  <span class="code-keyword">rank</span>:          16
  <span class="code-keyword">alpha</span>:         32
  <span class="code-keyword">dropout</span>:       0.05
  <span class="code-keyword">target_modules</span>: <span class="code-string">["q_proj","k_proj","v_proj","o_proj"]</span>

<span class="code-keyword">data</span>:
  <span class="code-keyword">corpus</span>:        <span class="code-string">sprintloop-corpus</span>
  <span class="code-keyword">examples</span>:      2800
  <span class="code-keyword">customer_data</span>: <span class="code-keyword">false</span>      <span class="code-comment"># enforced at registry</span>

<span class="code-keyword">train</span>:
  <span class="code-keyword">batch_size</span>:    4
  <span class="code-keyword">learning_rate</span>: <span class="code-number">1e-4</span>
  <span class="code-keyword">select_on</span>:     <span class="code-string">val_loss_min</span>   <span class="code-comment"># 0.69 @ iter 20</span>

<span class="code-function">register</span>(<span class="code-keyword">sign</span>=<span class="code-keyword">true</span>, <span class="code-keyword">channel</span>=<span class="code-string">"candidate"</span>)

Alignment

Sharpen behavior after the task is learned

Fine-tuning and alignment are documented design choices for a future governed runner. Their catalog status describes evidence or roadmap maturity, not executable Studio availability.

Reference

DPO / GRPO

preference + reasoning RL — sharpen refusal behavior

Roadmap

RLAIF / self-rewarding

AI feedback loop — scale alignment without labelers

What you get

An authoritative runner must end with defensible evidence

Before training can return as a supported lane, its output must bind exact weights, source authority, data, evaluation, and immutable provenance into a non-routable candidate.

A versioned adapter

A pinned artifact tagged to its exact base and recipe. Historical versions may remain inspectable, but no adapter is promised to be re-servable.

A tracked val-loss curve

The validation-loss minimum that selected the checkpoint, recorded against corpus size — the evidence that promotion was earned, not assumed.

A signed record

The run's lineage sealed into the registry and ready for the benchmark and release gates — no hand-off, no re-entry, no second source of truth.

SprintLoop-7B · v0 (POC)32B v1 · monitor: no regression · 14 cycles · next: harder 40–60 probe benchmark to separate 7B vs 32B

Review the training evidence contract

Inspect the roadmap and completed-run evidence requirements. Generic training remains unavailable until authoritative model, dataset, runner, and artifact identities are enforced.

View the training roadmapSee the full lifecycle