Vox MENS — Qwen3-0.6B adapter

0.6B is trained, its spoke is vox-lang and its gate is passed.

Why this size

The smallest rung, aimed at machines with very little memory, including CPU-only boxes. It suits quick completions and short single-function edits where reply time matters more than depth. Step up when it loses the thread of a multi-step task; there is no smaller size to step down to. The measured training peak for 0.6B is 19443 MiB (peak_source: run_summary), and no serving memory figure has been measured for it.

Examples

For 0.6B, the examples are the 2 shortest items the adapter fixes and the shortest it still fails; of 52 items it fixes 14 and breaks 3.

Fixed humaneval_vox_046_count_primes_up_to

Return the count of prime numbers in the range [2, n] inclusive. Return 0 if n < 2.

Qwen3-0.6B base on humaneval_vox_046_count_primes_up_to: does not parse (count 5).

fn count_primes(n: int) to int {
    if n < 2 {
        return 0
    }
    let count = 0
    for i in 2..n {
        if i % 2 == 0 {
            continue
        }
        if i % 3 == 0 {
            continue
        }
        if i % 5 == 0 {
            continue
        }
        count += 1
    }
    return count
}

0.6B adapter on humaneval_vox_046_count_primes_up_to: passes.

fn count_primes(n: int) to int {
    if n < 2 {
        return 0
    }
    let mut count = 0
    let mut max = n
    let mut i = 2
    while i <= max {
        if n % i == 0 {
            count = count + 1
        }
        i = i + 1
    }
    return count
}

Fixed page_index

define a page named `IndexPage` with a title 'Home' that displays a heading with the text 'Welcome'

Qwen3-0.6B base on page_index: does not parse (count 1).

page IndexPage {
    title: "Home"
    view: heading {
        text: "Welcome"
    }
}

0.6B adapter on page_index: passes.

component IndexPage() {
    state title: str = "Home"
    view: column() {
        heading(level=1) { "Welcome" }
    }
}

Still fails humaneval_vox_042_reverse_list

Return a new list with the elements in reverse order.

Qwen3-0.6B base on humaneval_vox_042_reverse_list: passes.

fn reverse_list(input: list) to list {
    return input
}

0.6B adapter on humaneval_vox_042_reverse_list: type errors (count 1).

fn reverse_list(items: list[int]) to list[int] {
    return items.reverse()
}

Use it

The 0.6B adapter sits on the base pin Qwen/Qwen3-0.6B@c1899de289a04d12100db370d81485cdf75e47ca, and this command downloads it after checking the BLAKE3 hash of every file in hf_sync_manifest.json.

vox mens hub pull --size 0.6B --namespace brbrainerd

The pull writes the 0.6B run directory that this command serves.

vox mens serve --model mens/runs/qwen3-0.6b --port 11435

This request sends the user turn of held-out item action_delete to the 0.6B server, which applies its codegen lane to it.

curl -s http://127.0.0.1:11435/v1/chat/completions -H 'Content-Type: application/json' -d @- <<'JSON'
{
  "lane": "codegen",
  "messages": [
    {
      "role": "user",
      "content": "Write a Vox program that: define an action named `delete_item` that takes an integer parameter `id` and returns a boolean indicating success"
    }
  ],
  "max_tokens": 1024,
  "temperature": 0.0
}
JSON

The code below is the 0.6B adapter's recorded output for item action_delete in the held-out evaluation, taken from evidence/heldout_adapter_nothink.ckpt.jsonl, where the codegen lane's prompt protocol qwen3_non_thinking_codegen_v1 wrapped the same user turn and every stored check passed; it is not a transcript of vox mens serve.

fn delete_item(id: int) to bool {
    return true
}

Results

The 0.6B rows compare the base model with the adapter on identical items, and a slice p-value below 0.05 would mean the adapter lost ground on that slice.

The release gate record for 0.6B is gate_receipt.json.

Training

0.6B ran 480 optimizer steps over 3447051 tokens, ended with a final loss (EMA) of 1.6497, a mean train loss of 1.9265 and a last-epoch validation loss of 1.5229, took 2h 50m and peaked at 19443 MiB, from Vox commit 36c233789; the settings every size shares are on the docs page.

Files

The 0.6B repository holds candle_qlora_adapter.safetensors, the adapter, training and gate manifests and records, evidence/, and hf_sync_manifest.json with the BLAKE3 hash of every file.

This card is generated by vox mens hub sync; the ladder index lists the other sizes and how the cards are made is on the docs page.