Downloads · 30 days
0
mcp-tool-shop/jam-actions-v1-qwen25-1.5b
jam-actions-v1-qwen25-1.5b is a machine learning model from mcp-tool-shop. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for peft. The card lists the license as cc-by-sa-3.0.
Two LoRA adapters, one recipe at two seeds, on Qwen2.5-1.5B-Instruct (Apache-2.0), trained on jam-actions-v1 1.1.0 (the four-draw corpus of the eleven verified songs). This is the smallest model that learned the acous…
Downloads · 30 days
0
Access
Public
Updated Sep 9, 2026
Repo size
159 MB
Likes
0
Public
Click a slice to open those files.
.safetensors148 MB · 87%
From the Hugging Face model README
Two LoRA adapters, one recipe at two seeds, on Qwen2.5-1.5B-Instruct (Apache-2.0), trained on jam-actions-v1 1.1.0 (the four-draw corpus of the eleven verified songs). This is the smallest model that learned the acoustic gate comparison from this corpus. It is an option for hardware where 4B is too much; the recommended small path is mcp-tool-shop/jam-actions-v1-qwen3-4b, and the 7B is at mcp-tool-shop/jam-actions-v1-qwen25-7b.
Every number below is on a held-out split by song or on the never-trained near-gate probe; the base model is reported on the same split; every seed that was run is reported.
bf16 LoRA, r = 16, α = 32, dropout 0.1, on q/k/v/o/gate/up/down; lr 1.5e-4 cosine, 10 warmup
steps; effective batch 8 (1 × 8 accumulation); three epochs; max_seq_len 16,384; prompt-loss
weight 0.1; chunked cross-entropy. Predictions greedy, max_new_tokens 128. Receipts beside each
adapter carry the seed, data SHA-256, hyperparameters, package versions and the loss curve.
| directory | seed | 1.1.0 held-out acoustic (36) | 1.0.0 held-out acoustic (17) | 1.0.0 overall (40) | near-gate probe (24) |
|---|---|---|---|---|---|
qwen25-1.5b-s13/ | 13 | 34/36 | 15/17 | 33/40 | 24/24 |
qwen25-1.5b-s42/ | 42 | 36/36 | 17/17 | 36/40 | 24/24 |
Base on the same splits: 12/36, 6/24. Read the shown work with care. Both seeds get the label right near the gate 24/24, but the subtraction they write is exact on only 19/24 (seed 13) and 10/24 (seed 42) of those lines: the word follows the model's own arithmetic, and that arithmetic is often off by tenths. On its own held-out takes seed 42 is exact on 31/36. This is a model that learned the decision better than the digits; the 4B and 7B write both correctly. What the 1.5B does and does not reach, seed 13: near the gate it is exact on 19 of 24 subtractions and right on all 24 labels; on its own held-out takes 34/36. Its held-out overall of 33/40 is the non-acoustic families — harmony 3/6 and key_moments 1/2 where the 4B has 6/6 and 1/2 — so it is a grader for the acoustic comparison, not for the rest of the tool surface.
They write the comparison and then the label:
cents 66.9: |66.9| − 50 = 16.9, against the gate; onset −9.8: |9.8| − 40 = −30.2, inside: pitch_fail
They were trained to compare against these gates (50 cents, 40 ms, chromatic ratio 0.2) in this tool vocabulary, on eleven songs. They are not general graders.
The adapters are released under CC-BY-SA-3.0-DE, matching the dataset they were trained on. Use of a merged or adapted model is additionally governed by the base model's licence, and here that is Apache-2.0.
Source repo experiments/coverage-v1-sft/: scripts/train_v1_sft.py, scripts/predict_v1.py,
scripts/score_v1.mjs, lora-config-1.5b.json, RESULTS-r52.md; the corpus is
datasets/jam-actions-v1 at 1.1.0 (Zenodo 10.5281/zenodo.22679457).