Add current v4/v5 LM sliders (structured captions + end regularizer)

Ship the studio's current language-model halves: gender/rapslow/triphop/energy/tempo/distortion/breath-lm-v4, live-lm-v5, rhyme-lm-v5. Add v4-endreg-ab-90s, live-endreg A/Bs, and triphop-v4-unit samples. v3 weights stay for comparison.

.gitattributes

@@ -208,3 +208,22 @@ samples/rhyme-v3-lm-20s/04_slider_Rhyme_plus1.wav filter=lfs diff=lfs merge=lfs
 samples/rhyme-v3-lm-20s/05_slider_Rhyme_plus2.wav filter=lfs diff=lfs merge=lfs -text
 samples/rhyme-v3-lm-20s/06_REF_prompt_Rhyme_no_slider.wav filter=lfs diff=lfs merge=lfs -text
 samples/rhyme-v3-lm-20s/07_REF_prompt_Prose_no_slider.wav filter=lfs diff=lfs merge=lfs -text
+samples/live-endreg-ab-60s/01_base.wav filter=lfs diff=lfs merge=lfs -text
+samples/live-endreg-ab-60s/02_live-v3+2.wav filter=lfs diff=lfs merge=lfs -text
+samples/live-endreg-ab-60s/03_live-v4+2.wav filter=lfs diff=lfs merge=lfs -text
+samples/live-endreg-ab-90s/01_base.wav filter=lfs diff=lfs merge=lfs -text
+samples/live-endreg-ab-90s/02_live-v3+2.wav filter=lfs diff=lfs merge=lfs -text
+samples/live-endreg-ab-90s/03_live-v4+2.wav filter=lfs diff=lfs merge=lfs -text
+samples/triphop-v4-unit-20s/01_shipped_Pop_minus2.wav filter=lfs diff=lfs merge=lfs -text
+samples/triphop-v4-unit-20s/02_shipped_zero.wav filter=lfs diff=lfs merge=lfs -text
+samples/triphop-v4-unit-20s/03_shipped_Trip-hop_plus2.wav filter=lfs diff=lfs merge=lfs -text
+samples/v4-endreg-ab-90s/01_base.wav filter=lfs diff=lfs merge=lfs -text
+samples/v4-endreg-ab-90s/02_gender-v4+2.wav filter=lfs diff=lfs merge=lfs -text
+samples/v4-endreg-ab-90s/03_rapslow-v4+2.wav filter=lfs diff=lfs merge=lfs -text
+samples/v4-endreg-ab-90s/04_triphop-v4+2.wav filter=lfs diff=lfs merge=lfs -text
+samples/v4-endreg-ab-90s/05_energy-v4+2.wav filter=lfs diff=lfs merge=lfs -text
+samples/v4-endreg-ab-90s/06_tempo-v4+2.wav filter=lfs diff=lfs merge=lfs -text
+samples/v4-endreg-ab-90s/07_distortion-v4+2.wav filter=lfs diff=lfs merge=lfs -text
+samples/v4-endreg-ab-90s/08_breath-v4+2.wav filter=lfs diff=lfs merge=lfs -text
+samples/v4-endreg-ab-90s/09_live-v5+2.wav filter=lfs diff=lfs merge=lfs -text
+samples/v4-endreg-ab-90s/10_rhyme-v5+2.wav filter=lfs diff=lfs merge=lfs -text

README.md

@@ -23,6 +23,10 @@ Training code lives in
 This repository holds the trained weights and the rendered listening examples,
 which are too large to keep in git.
 
+**Current language-model halves are v4 (live and rhyme are v5).** They add
+Structured Caption poles and an audio-end regularizer on top of the v3 recipe.
+Earlier v3 files stay in `weights/` for comparison.
+
 ## Which stage a slider attaches to
 
 MiniMax Music 3 generates in two stages, and this turns out to decide where a
@@ -39,20 +43,20 @@ performance* has to be trained on the language model:
 
 | slider | poles | host | notes |
 |---|---|---|---|
-| `gender-lm-v3` | male ↔ female | language model | a transformer version moved median F0 by 1 Hz — see below |
-| `rapslow-lm-v3` | slow sung ↔ rap flow | language model | transformer version failed the same way |
-| `triphop-lm-v3` | glossy pop ↔ trip-hop | language model | pairs with `triphop-slider` |
-| `energy-lm-v3` | quiet ↔ loud | language model | changes how hard the band *plays* |
-| `tempo-lm-v3` | slow ↔ fast | language model | changes the composed BPM |
-| `distortion-lm-v3` | clean ↔ heavy | language model | changes instrumentation |
+| `gender-lm-v4` | male ↔ female | language model | a transformer version moved median F0 by 1 Hz |
+| `rapslow-lm-v4` | slow sung ↔ rap flow | language model | transformer version failed the same way |
+| `triphop-lm-v4` | glossy pop ↔ trip-hop | language model | pairs with `triphop-tf-v4` |
+| `energy-lm-v4` | quiet ↔ loud | language model | changes how hard the band *plays* |
+| `tempo-lm-v4` | slow ↔ fast | language model | changes the composed BPM |
+| `distortion-lm-v4` | clean ↔ heavy | language model | changes instrumentation |
+| `breath-lm-v4` | clean ↔ breathy | language model | breath, mouth air, inhales |
+| `live-lm-v5` | studio ↔ live | language model | room, crowd, performance energy |
+| `rhyme-lm-v5` | prose ↔ rhyme | language model | lyric delivery; structured captions + endreg |
 | `energy-slider-v2` | quiet ↔ loud | transformer | mix-level loudness |
 | `distortion-slider` | acoustic ↔ metal | transformer | tone |
 | `tempo-slider` | slow ↔ fast | transformer | perceived pulse |
 | `space-slider` | dry ↔ wet | transformer | reverb — the one axis that is transformer-only |
-| `triphop-tf-v4` | pop ↔ trip-hop | transformer | style/production — clean retrain, ships as `_unit_last` |
-| `breath-lm-v3` | clean ↔ breathy | language model | breath, mouth air, inhales |
-| `live-lm-v3` | studio ↔ live | language model | room, crowd, performance energy |
-| `rhyme-lm-v4` | prose ↔ rhyme | language model | lyric delivery; trained on structured captions |
+| `triphop-tf-v4` | pop ↔ trip-hop | transformer | style/production — ships as `_unit_last` |
 
 Several axes work best with **both** halves applied together (transformer for the
 mix, language model for the arrangement). `samples/energy-v3-shipped-20s/` is an
@@ -67,17 +71,22 @@ good; a high value means both poles shift the same shared way):
 
 | axis | sep | cos | verdict |
 |---|---|---|---|
-| rap ↔ slow | 0.32 | 0.04 | strong |
-| energy | 0.28 | −0.10 | strong |
-| distortion | 0.23 | 0.03 | usable |
-| tempo | 0.21 | 0.34 | usable |
-| gender | 0.20 | 0.03 | usable — and it demonstrably works, so ~0.2 is the floor |
-| trip-hop | 0.20 | 0.28 | usable |
+| rap ↔ slow | 0.34 | −0.08 | strong |
+| energy | 0.33 | 0.03 | strong |
+| tempo | 0.31 | 0.06 | strong |
+| rhyme | 0.27 | −0.05 | usable |
+| distortion | 0.27 | −0.03 | usable |
+| breath | 0.27 | 0.24 | usable |
+| live | 0.25 | 0.32 | usable |
+| gender | 0.24 | −0.08 | usable — and it demonstrably works, so ~0.2 is the floor |
+| trip-hop | 0.22 | 0.10 | usable |
 | space (dry ↔ wet) | 0.17 | **0.70** | transformer only |
 
 Space is the informative failure: its two poles move the plan mostly in the *same*
 direction, i.e. the model encodes "this caption is about room acoustics" rather
-than an opposing dry/wet axis.
+than an opposing dry/wet axis. The table above is from the v4 Structured Caption
+poles (v3 numbers were similar; tempo's shared-component cos dropped from 0.34
+to 0.06).
 
 ## Usage
 
@@ -95,7 +104,7 @@ from conceptmod.textsliders.lora import LoRANetwork
 pipe = ModularPipeline.from_pretrained("path/to/MiniMax-Music3")
 # ... load components, move to device ...
 
-weights = Path("weights/gender-lm-v3/gender-lm-v3_last.safetensors")
+weights = Path("weights/gender-lm-v4/gender-lm-v4_last.safetensors")
 meta = json.loads(weights.with_suffix(".json").read_text())
 host = pipe.language_model if meta["kind"] == "language_model" else pipe.transformer
 
@@ -139,21 +148,44 @@ sound to compare against. Each folder has a `LISTEN.md` with a play order and mo
 have a `probe.json` of measurements.
 
 Start with `samples/energy-v3-lm-20s/`, `samples/gender-v3-ballad-20s/` and
-`samples/triphop-v3-stack-20s/`.
+`samples/triphop-v3-stack-20s/`. For the current LM halves, play
+`samples/v4-endreg-ab-90s/` (same-seed 90s +2 of every shipped LM slider vs base).
 
 ## August 2026 update
 
+- **LM halves retrained as v4 / v5.** Every language-model slider was retrained
+  with Structured Caption poles (the format the studio rewriter actually emits)
+  and an **audio-end regularizer**. Un-regularized LM LoRAs make
+  `<|audio_end|>` less likely, so long renders hit the duration cap mid-phrase.
+  Training now teacher-forces the end-margin from a frozen base-model
+  composition. `samples/v4-endreg-ab-90s/` is the same-seed 90s check: eight of
+  nine sliders at +2 end naturally; rhyme-v5 hit the cap on that one seed.
+  `samples/live-endreg-ab-90s/` is the v3-vs-v4 ending A/B on the live axis.
+  Final-window metrics (collapse / perc / edrift):
+
+  | checkpoint | collapse | perc | edrift |
+  |---|--:|--:|--:|
+  | `gender-lm-v4` | −0.95 | 0.23 | 0.060 |
+  | `rapslow-lm-v4` | −0.96 | 0.29 | 0.080 |
+  | `triphop-lm-v4` | −0.95 | 0.25 | 0.059 |
+  | `energy-lm-v4` | −0.97 | 0.29 | 0.072 |
+  | `tempo-lm-v4` | −0.95 | 0.30 | 0.071 |
+  | `distortion-lm-v4` | −0.96 | 0.33 | 0.076 |
+  | `breath-lm-v4` | −0.95 | 0.36 | 0.069 |
+  | `live-lm-v5` | −0.91 | 0.43 | 0.068 |
+  | `rhyme-lm-v5` | −0.95 | 0.34 | 0.070 |
+
+  All nine are `unit_scale` 1.0. v3 LM files (and `rhyme-lm-v4`) remain in
+  `weights/` for comparison; they do not carry the end regularizer.
 - **`triphop-slider` is replaced by `triphop-tf-v4`.** The old transformer half was
   trained on a prompt that named real bands; it has been retrained from an
   equivalent artist-free prompt (matching metrics: loss 0.090, pole opposition
   −0.89) and removed from this repository. The new file ships unit-normalized.
-- **New rhyme axis.** `rhyme-lm-v4` is trained on poles written in the model's
-  official Structured Caption format (Global Metadata / Vocal Details /
-  Arrangement), carrying the axis in Vocal Details + Arrangement — on-distribution
-  with runtime captions. It beats the flat-caption v3 on every axis metric
-  (pole opposition −0.992 vs −0.974; probe sep 0.270 vs 0.248). `rhyme-lm-v3`
-  is included for comparison. `samples/rhyme-v4-caption-pair/` shows it behaving
-  on an unseen full structured caption.
+- **New rhyme axis.** `rhyme-lm-v5` is the current file: structured-caption poles
+  plus the end regularizer. `rhyme-lm-v4` (structured captions, no endreg) and
+  `rhyme-lm-v3` (flat captions) stay for comparison.
+  `samples/rhyme-v4-caption-pair/` shows the structured-caption version on an
+  unseen full caption.
 - **Stacking measured.** `samples/stack-sweep-20s/` renders escalating slider
   stacks up to all ten sliders at ±2 (combined total 27 units): rms stays within
   0.7–2.0× of baseline and crest factor never collapses — no degradation knee was
@@ -171,9 +203,12 @@ Start with `samples/energy-v3-lm-20s/`, `samples/gender-v3-ballad-20s/` and
   `triphop-ab-anchor-single` are ablation runs, not sliders to use.
 - **Loudness entanglement.** Transformer sliders pushed past roughly raw ±4 tend to
   slam or hollow out the mix rather than express the concept.
+- **Endings are sampled.** The end regularizer makes a natural fade much more
+  likely, but it is not a guarantee on every seed. rhyme-v5 +2 hit the 90s cap
+  on seed 7 in `samples/v4-endreg-ab-90s/`.
 - Trained at 4-second conditioning; generation length is independent of that.
 
-## Training recipe (v3)
+## Training recipe (v3, still the base)
 
 Rank 8, alpha 8, 500 steps at lr 1e-4 for transformer sliders; 800 steps at lr
 5e-4 for language-model sliders. Four fixes distinguish v3 from earlier weights:
@@ -188,6 +223,8 @@ Rank 8, alpha 8, 500 steps at lr 1e-4 for transformer sliders; 800 steps at lr
 4. **Module-identity dedupe** in the LoRA wrapper — targeting the model root used
    to wrap every attention/FF linear twice, at double gain.
 
+v4/v5 keep that recipe and add Structured Caption poles plus `--endreg_weight 1.0`.
+
 For prompt writing: language-model sliders want *divergent* poles (let genre, BPM
 and instrumentation all move). Rewriting energy as one swapped clause in an
 otherwise identical caption dropped separation from 0.30 to 0.11.

samples/README.md

@@ -5,6 +5,18 @@ Same lyrics (`I can feel it in the air tonight / Louder now or fade away`) and s
 - **Slider clips** use the neutral caption. Only the LoRA scale changes.
 - **REF clips** change the prompt with the slider off. That is the target sound.
 
+**Current shipped language-model halves are v4 (live and rhyme are v5).**
+Those files live under `weights/*-lm-v4/` and `weights/{live,rhyme}-lm-v5/`.
+The 20s axis folders below are still the v3 renders.
+
+## Current LM halves (v4/v5)
+
+| folder | what |
+|---|---|
+| `v4-endreg-ab-90s/` | same-seed 90s +2 of every shipped LM half vs base — ending check |
+| `live-endreg-ab-60s/` / `live-endreg-ab-90s/` | live-v3 vs live-v4 ending A/B (un-regularized v3 hits the cap) |
+| `triphop-v4-unit-20s/` | unit-normalized transformer half as the studio plays it |
+
 Start with the **20s** folders. Those are the end-to-end long renders. `energy-30s/` is a longer-duration check of the loudness slider. The 8s folders are the earlier smoke tests (energy, distortion, gender-lm).
 
 **Gender identity (24s, several methods):** `gender-24s/`. Transformer LoRA does not change F0; the LM slider and the condition rewrite do. Play `08`/`09` then `02`/`03`.
@@ -31,8 +43,8 @@ Each folder also has `probe.json` from `scripts/probe_axis.py`.
 | `rapslow-v3-tf-20s/` | delivery (TF) | — | — | **negative result**: TF calibration cos_axis ≈ 0.08, unit_scale blew up to 6.7, clips overdriven. Like gender, delivery is decided in the AR LM — use the LM slider. |
 | `triphop-v3-lm-20s/` | style (LM v3) | glossy pop | trip-hop | passes ref-relative centroid gate (63% of ref span) |
 | `triphop-v3-tf-20s/` | style (TF, multi-row) | — | — | **negative result**: multi-row averaging kills TF axis tracking (cos_axis 0.05) |
-| `triphop-v3-tf-single-20s/` + `triphop-v3-tf-raw-20s/` | style (TF v3, anchor+single) | — | — | trains better (cos 0.51) but entangles loudness: even raw ±2 slams (−) or silences (+) the mix. The app keeps **TF v2** for triphop. |
-| `triphop-v3-stack-20s/` | style (TF+LM stacked) | pop | trip-hop | the app's actual `triphop` slider configuration |
+| `triphop-v3-tf-single-20s/` + `triphop-v3-tf-raw-20s/` | style (TF v3, anchor+single) | — | — | trains better (cos 0.51) but entangles loudness: even raw ±2 slams (−) or silences (+) the mix. The app keeps **TF v4** for triphop. |
+| `triphop-v3-stack-20s/` | style (TF+LM stacked) | pop | trip-hop | stacked configuration |
 | `energy-v3-lm-20s/` | energy (LM v3) | quiet | loud | **PASS**: rms rises monotonically 0.075 → 0.133, +2 lands on the loud REF |
 | `energy-v3-stack-20s/` | energy (TF+LM) | quiet | loud | A/B: LM-only lands on the loud REF; the TF half at its calibrated unit (raw 4.02) overshoots |
 | `energy-v3-shipped-20s/` | energy (TF+LM) | quiet | loud | **what the studio actually plays** at user ±2, via `scripts/render_shipped_slider.py` |
@@ -77,17 +89,24 @@ If a slider works, clip 3 leans toward clip 4 and clip 1 leans toward clip 5.
 
 ## Weights
 
-- `models/energy-slider-v2/energy_alpha8.0_rank8_full_last.safetensors`
-- `models/distortion-slider/distortion_alpha8.0_rank8_full_last.safetensors`
-- `models/tempo-slider/tempo_alpha8.0_rank8_full_last.safetensors`
-- `models/space-slider/space_alpha8.0_rank8_full_last.safetensors`
-- `models/gender-lm-slider/gender-lm_last.safetensors` (`--kind lm`)
+Current language-model halves:
+
+- `weights/gender-lm-v4/gender-lm-v4_last.safetensors`
+- `weights/rapslow-lm-v4/rapslow-lm-v4_last.safetensors`
+- `weights/triphop-lm-v4/triphop-lm-v4_last.safetensors`
+- `weights/energy-lm-v4/energy-lm-v4_last.safetensors`
+- `weights/tempo-lm-v4/tempo-lm-v4_last.safetensors`
+- `weights/distortion-lm-v4/distortion-lm-v4_last.safetensors`
+- `weights/breath-lm-v4/breath-lm-v4_last.safetensors`
+- `weights/live-lm-v5/live-lm-v5_last.safetensors`
+- `weights/rhyme-lm-v5/rhyme-lm-v5_last.safetensors`
 
-Transformer sliders: rank-8 LoRA on `MiniMaxMusic3Attention`, 500 steps. Gender: rank-8 LoRA on `Qwen3Attention`, 800 steps. All trained and rendered on GPU 0.
+Transformer halves (unchanged):
 
-Regenerate or verify:
+- `weights/energy-slider-v2/energy_unit_last.safetensors`
+- `weights/distortion-slider/distortion_alpha8.0_rank8_full_last.safetensors`
+- `weights/tempo-slider/tempo_alpha8.0_rank8_full_last.safetensors`
+- `weights/space-slider/space_alpha8.0_rank8_full_last.safetensors`
+- `weights/triphop-tf-v4/triphop-tf-v4_unit_last.safetensors`
 
-```bash
-./scripts/run_gpu0_sliders.sh demo-20
-./scripts/run_gpu0_sliders.sh verify
-```
+v3 LM files stay in `weights/` for comparison.

samples/live-endreg-ab-60s/01_base.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:07cdda2f1ebb603e7de6df6de34b528b43aeadef288067feae0bd9ef3d276f42
+size 10596396

samples/live-endreg-ab-60s/02_live-v3+2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b1e07cb75903ce0983acd8ca869eeae2325b5a2b0f6e0f32b257d000ca94db1
+size 10596396

samples/live-endreg-ab-60s/03_live-v4+2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f5293df1db41878d7afbe51e7c848b39d01259eb22b95ec7a8e3e8ff7b2b5c3
+size 10596396

samples/live-endreg-ab-60s/LISTEN.md

@@ -0,0 +1,12 @@
+# Same-seed ending A/B
+
+Source: library song `5ec87fed`. Seed 7, requested 60s — identical for every row,
+so the only difference is the applied LoRA. tail/overall well under 0.1 means the cut
+faded out on its own; a hot tail at the cap means the composer never sampled
+`<|audio_end|>` and was guillotined.
+
+| file | scale | seconds | tail/overall | ended naturally |
+|------|------:|--------:|-------------:|-----------------|
+| `01_base.wav` | +0 | 60.07 | 0.147 | **no — cap** |
+| `02_live-v3+2.wav` | +2 | 60.07 | 0.936 | **no — cap** |
+| `03_live-v4+2.wav` | +2 | 60.07 | 0.436 | **no — cap** |

samples/live-endreg-ab-90s/01_base.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80d5b3b53c5ea8b049316e5f3385d8590e795d92fe5b5a945724a53103ab18ec
+size 13535276

samples/live-endreg-ab-90s/02_live-v3+2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0229df90cae2f25e79de4a2f20250ba1340aab550dad693ff96cd7781d749fd0
+size 15894572

samples/live-endreg-ab-90s/03_live-v4+2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4347b7aa2ba8387965bacd473ccbf8a5f94b4a2e623241b515525250d623e2e0
+size 14991404

samples/live-endreg-ab-90s/LISTEN.md

@@ -0,0 +1,12 @@
+# Same-seed ending A/B
+
+Source: library song `5ec87fed`. Seed 7, requested 90s — identical for every row,
+so the only difference is the applied LoRA. tail/overall well under 0.1 means the cut
+faded out on its own; a hot tail at the cap means the composer never sampled
+`<|audio_end|>` and was guillotined.
+
+| file | scale | seconds | tail/overall | ended naturally |
+|------|------:|--------:|-------------:|-----------------|
+| `01_base.wav` | +0 | 76.73 | 0.020 | yes |
+| `02_live-v3+2.wav` | +2 | 90.11 | 1.045 | **no — cap** |
+| `03_live-v4+2.wav` | +2 | 84.99 | 0.018 | yes |

samples/triphop-v4-unit-20s/01_shipped_Pop_minus2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99a3ddc3c067d41a2ac33ed98fcab20b968db96981635d38aecc5f22003247d2
+size 3530796

samples/triphop-v4-unit-20s/02_shipped_zero.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9bf3b5d0e031977f1a2d1758c68ada28b77a48c722c93a4980eeda2b94be3be7
+size 3530796

samples/triphop-v4-unit-20s/03_shipped_Trip-hop_plus2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b61f76c88c4eef404cbf0fb39baf4b9575e00cc0d5fe5e7448d606fadad6b1a
+size 3530796

samples/triphop-v4-unit-20s/LISTEN.md

@@ -0,0 +1,13 @@
+# triphop — exactly what the studio ships
+
+Resolved through `app/sliders.json` (Pop <-> Trip-hop), so the multipliers
+below include each component's ratio, gain and sidecar `unit_scale`.
+Same neutral caption, lyrics and seed throughout.
+
+| file | seconds | rms | applied |
+|------|--------:|----:|---------|
+| `01_shipped_Pop_minus2.wav` | 20.02 | 0.1474 | triphop-tf-v4x-2.00, triphop-lm-v3x-2.00 |
+| `02_shipped_zero.wav` | 20.02 | 0.0975 | none |
+| `03_shipped_Trip-hop_plus2.wav` | 20.02 | 0.0718 | triphop-tf-v4x+2.00, triphop-lm-v3x+2.00 |
+
+- seed: 7  duration: 20.0s  prompt row: 0

samples/v4-endreg-ab-90s/01_base.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80d5b3b53c5ea8b049316e5f3385d8590e795d92fe5b5a945724a53103ab18ec
+size 13535276

samples/v4-endreg-ab-90s/02_gender-v4+2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:36bf73049215eaf0c744e29dd96279344b888faaad8668fed2592eae0d108428
+size 14377004

samples/v4-endreg-ab-90s/03_rapslow-v4+2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2cac56e812dc470bdbebd1dc08aab8a178f85793f759b119a934525e8182763e
+size 14608428

samples/v4-endreg-ab-90s/04_triphop-v4+2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:56ee099fa1c7e511f9eddd2b7413f7d2f3697b230f14a6a951875a3b73a220fd
+size 15054892

samples/v4-endreg-ab-90s/05_energy-v4+2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6c764f00cb384e41afa1182e38b851f78b3a3fb8177bb0704a5204c180d57ab
+size 12808236

samples/v4-endreg-ab-90s/06_tempo-v4+2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:62a4b5b2b27f61470c5251f805a333a59b46becb81b1e65efb1e1ab36a71ab15
+size 11501612

samples/v4-endreg-ab-90s/07_distortion-v4+2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:056441f028391015f207e7512a8e0e954cce50ee59b1ba184abfe1daad91e73e
+size 14581804

samples/v4-endreg-ab-90s/08_breath-v4+2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f5145233ee90cb25bfe321432f3c85a03f9e2afa3911525f5e234135aff6bb9
+size 10971180

samples/v4-endreg-ab-90s/09_live-v5+2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6eba7b26782e909896655953913c94f473d51c2f87487b0bac807b07c91be10e
+size 11952172

samples/v4-endreg-ab-90s/10_rhyme-v5+2.wav

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c792ba62405c6b762a9f3fed9ae9caf3ad2fe972234a6d34173daf62a1badf9
+size 15894572

samples/v4-endreg-ab-90s/LISTEN.md

@@ -0,0 +1,19 @@
+# Same-seed ending A/B
+
+Source: library song `5ec87fed`. Seed 7, requested 90s — identical for every row,
+so the only difference is the applied LoRA. tail/overall well under 0.1 means the cut
+faded out on its own; a hot tail at the cap means the composer never sampled
+`<|audio_end|>` and was guillotined.
+
+| file | scale | seconds | tail/overall | ended naturally |
+|------|------:|--------:|-------------:|-----------------|
+| `01_base.wav` | +0 | 76.73 | 0.020 | yes |
+| `02_gender-v4+2.wav` | +2 | 81.50 | 0.025 | yes |
+| `03_rapslow-v4+2.wav` | +2 | 82.81 | 0.019 | yes |
+| `04_triphop-v4+2.wav` | +2 | 85.34 | 0.276 | yes |
+| `05_energy-v4+2.wav` | +2 | 72.61 | 0.024 | yes |
+| `06_tempo-v4+2.wav` | +2 | 65.20 | 0.428 | yes |
+| `07_distortion-v4+2.wav` | +2 | 82.66 | 0.658 | yes |
+| `08_breath-v4+2.wav` | +2 | 62.19 | 0.023 | yes |
+| `09_live-v5+2.wav` | +2 | 67.76 | 0.018 | yes |
+| `10_rhyme-v5+2.wav` | +2 | 90.11 | 1.644 | **no — cap** |

weights/breath-lm-v4/breath-lm-v4_last.json

@@ -0,0 +1,79 @@
+{
+  "schema": 3,
+  "name": "breath-lm-v4",
+  "checkpoint": "/ml2/music/sliders-conceptmod/models/breath-lm-v4/breath-lm-v4_last.safetensors",
+  "weights": "/ml2/music/sliders-conceptmod/models/breath-lm-v4/breath-lm-v4_last.safetensors",
+  "modules": 144,
+  "rank": 8,
+  "alpha": 8.0,
+  "steps": 800,
+  "steps_budget": 800,
+  "lr": 0.0005,
+  "rows": 6,
+  "symmetric": true,
+  "common_beta": 0.0,
+  "first": {
+    "step": 1,
+    "row": 0,
+    "loss": 0.04791141301393509,
+    "pperc": 1.000883936882019,
+    "nperc": 1.005846619606018,
+    "cos_pos": 0.030307313427329063,
+    "cos_neg": -0.030307313427329063,
+    "collapse": 1.0000001192092896,
+    "edrift_p": 0.0,
+    "edrift_n": 0.0
+  },
+  "last": {
+    "step": 800,
+    "row": 1,
+    "loss": 0.011750648729503155,
+    "pperc": 0.31941771507263184,
+    "nperc": 0.2995215058326721,
+    "cos_pos": 0.9477377533912659,
+    "cos_neg": 0.9541085958480835,
+    "collapse": -0.9564054012298584,
+    "edrift_p": 0.06156808137893677,
+    "edrift_n": 0.06700307875871658
+  },
+  "target_replace": [
+    "Qwen3Attention"
+  ],
+  "kind": "language_model",
+  "prefix": "lora_te",
+  "delimiter": "-",
+  "train_method": "full",
+  "unit_scale": 1.0,
+  "endreg": {
+    "enabled": true,
+    "weight": 1.0,
+    "frames": 250,
+    "seed": 7,
+    "rows_ended_naturally": 0,
+    "final_edrift_p": 0.06906526900827885,
+    "final_edrift_n": 0.06836814865469933
+  },
+  "plus_label": "Breathy",
+  "minus_label": "Clean",
+  "recommended_range": [
+    -2.0,
+    2.0
+  ],
+  "prompts_file": "conceptmod/textsliders/data/prompts-breath-v4.yaml",
+  "early_stop": {
+    "enabled": true,
+    "fired": false,
+    "window": 50,
+    "min_steps": 100,
+    "cos": 0.97,
+    "collapse": -0.95,
+    "perc": 0.2,
+    "metrics": {
+      "cos_pos": 0.9378063213825226,
+      "cos_neg": 0.9309412884712219,
+      "collapse": -0.9482558608055115,
+      "pperc": 0.3490370923280716,
+      "nperc": 0.3624880576133728
+    }
+  }
+}

weights/breath-lm-v4/breath-lm-v4_last.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b311e073485d691704c1a41c0198ac9a54d685bfff1807408a43d6e58d94d6eb
+size 30723552

weights/distortion-lm-v4/distortion-lm-v4_last.json

@@ -0,0 +1,79 @@
+{
+  "schema": 3,
+  "name": "distortion-lm-v4",
+  "checkpoint": "/ml2/music/sliders-conceptmod/models/distortion-lm-v4/distortion-lm-v4_last.safetensors",
+  "weights": "/ml2/music/sliders-conceptmod/models/distortion-lm-v4/distortion-lm-v4_last.safetensors",
+  "modules": 144,
+  "rank": 8,
+  "alpha": 8.0,
+  "steps": 800,
+  "steps_budget": 800,
+  "lr": 0.0005,
+  "rows": 6,
+  "symmetric": true,
+  "common_beta": 0.0,
+  "first": {
+    "step": 1,
+    "row": 0,
+    "loss": 0.042385660111904144,
+    "pperc": 1.0102059841156006,
+    "nperc": 0.9951561093330383,
+    "cos_pos": -0.10243198275566101,
+    "cos_neg": 0.10243198275566101,
+    "collapse": 1.0000001192092896,
+    "edrift_p": 0.0,
+    "edrift_n": 0.0
+  },
+  "last": {
+    "step": 800,
+    "row": 1,
+    "loss": 0.01722337119281292,
+    "pperc": 0.3111982047557831,
+    "nperc": 0.2874501645565033,
+    "cos_pos": 0.9503713846206665,
+    "cos_neg": 0.957816481590271,
+    "collapse": -0.9676225185394287,
+    "edrift_p": 0.09480186551809311,
+    "edrift_n": 0.09529199451208115
+  },
+  "target_replace": [
+    "Qwen3Attention"
+  ],
+  "kind": "language_model",
+  "prefix": "lora_te",
+  "delimiter": "-",
+  "train_method": "full",
+  "unit_scale": 1.0,
+  "endreg": {
+    "enabled": true,
+    "weight": 1.0,
+    "frames": 250,
+    "seed": 7,
+    "rows_ended_naturally": 0,
+    "final_edrift_p": 0.0755588511377573,
+    "final_edrift_n": 0.075579244941473
+  },
+  "plus_label": "Heavy",
+  "minus_label": "Clean",
+  "recommended_range": [
+    -2.0,
+    2.0
+  ],
+  "prompts_file": "conceptmod/textsliders/data/prompts-distortion-v4.yaml",
+  "early_stop": {
+    "enabled": true,
+    "fired": false,
+    "window": 50,
+    "min_steps": 100,
+    "cos": 0.97,
+    "collapse": -0.95,
+    "perc": 0.2,
+    "metrics": {
+      "cos_pos": 0.9430427193641663,
+      "cos_neg": 0.9423296689987183,
+      "collapse": -0.9575636041164398,
+      "pperc": 0.3330520153045654,
+      "nperc": 0.3347309058904648
+    }
+  }
+}

weights/distortion-lm-v4/distortion-lm-v4_last.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a5488ff17e0b46d2d38f8c1edeade03147c417cd3333454ba6e8644cc6bb9971
+size 30723552

weights/energy-lm-v4/energy-lm-v4_last.json

@@ -0,0 +1,79 @@
+{
+  "schema": 3,
+  "name": "energy-lm-v4",
+  "checkpoint": "/ml2/music/sliders-conceptmod/models/energy-lm-v4/energy-lm-v4_last.safetensors",
+  "weights": "/ml2/music/sliders-conceptmod/models/energy-lm-v4/energy-lm-v4_last.safetensors",
+  "modules": 144,
+  "rank": 8,
+  "alpha": 8.0,
+  "steps": 800,
+  "steps_budget": 800,
+  "lr": 0.0005,
+  "rows": 6,
+  "symmetric": true,
+  "common_beta": 0.0,
+  "first": {
+    "step": 1,
+    "row": 0,
+    "loss": 0.04687363654375076,
+    "pperc": 1.0071539878845215,
+    "nperc": 1.0043737888336182,
+    "cos_pos": -0.013001742772758007,
+    "cos_neg": 0.013001742772758007,
+    "collapse": 1.0,
+    "edrift_p": 0.0,
+    "edrift_n": 0.0
+  },
+  "last": {
+    "step": 800,
+    "row": 1,
+    "loss": 0.013388051651418209,
+    "pperc": 0.3098142743110657,
+    "nperc": 0.29809361696243286,
+    "cos_pos": 0.9512045383453369,
+    "cos_neg": 0.9546085596084595,
+    "collapse": -0.9653064608573914,
+    "edrift_p": 0.07202227413654327,
+    "edrift_n": 0.07248590141534805
+  },
+  "target_replace": [
+    "Qwen3Attention"
+  ],
+  "kind": "language_model",
+  "prefix": "lora_te",
+  "delimiter": "-",
+  "train_method": "full",
+  "unit_scale": 1.0,
+  "endreg": {
+    "enabled": true,
+    "weight": 1.0,
+    "frames": 250,
+    "seed": 7,
+    "rows_ended_naturally": 0,
+    "final_edrift_p": 0.07162492625415325,
+    "final_edrift_n": 0.07267352074384689
+  },
+  "plus_label": "Loud",
+  "minus_label": "Quiet",
+  "recommended_range": [
+    -2.0,
+    2.0
+  ],
+  "prompts_file": "conceptmod/textsliders/data/prompts-energy-v4.yaml",
+  "early_stop": {
+    "enabled": true,
+    "fired": false,
+    "window": 50,
+    "min_steps": 100,
+    "cos": 0.97,
+    "collapse": -0.95,
+    "perc": 0.2,
+    "metrics": {
+      "cos_pos": 0.9590405368804932,
+      "cos_neg": 0.9564559578895568,
+      "collapse": -0.9658816075325012,
+      "pperc": 0.2830392503738403,
+      "nperc": 0.2934559726715088
+    }
+  }
+}

weights/energy-lm-v4/energy-lm-v4_last.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b0ae66dc96002ff7e1d7895b7ac55ab0e2e3686619329f3c2df7e07894b74e8a
+size 30723552

weights/gender-lm-v4/gender-lm-v4_last.json

@@ -0,0 +1,79 @@
+{
+  "schema": 3,
+  "name": "gender-lm-v4",
+  "checkpoint": "/ml2/music/sliders-conceptmod/models/gender-lm-v4/gender-lm-v4_last.safetensors",
+  "weights": "/ml2/music/sliders-conceptmod/models/gender-lm-v4/gender-lm-v4_last.safetensors",
+  "modules": 144,
+  "rank": 8,
+  "alpha": 8.0,
+  "steps": 800,
+  "steps_budget": 800,
+  "lr": 0.0005,
+  "rows": 4,
+  "symmetric": true,
+  "common_beta": 0.0,
+  "first": {
+    "step": 1,
+    "row": 0,
+    "loss": 0.031037069857120514,
+    "pperc": 0.987704336643219,
+    "nperc": 1.0218567848205566,
+    "cos_pos": 0.17265677452087402,
+    "cos_neg": -0.17265677452087402,
+    "collapse": 0.9999998807907104,
+    "edrift_p": 0.0,
+    "edrift_n": 0.0
+  },
+  "last": {
+    "step": 800,
+    "row": 3,
+    "loss": 0.0111676761880517,
+    "pperc": 0.2769600749015808,
+    "nperc": 0.2805417478084564,
+    "cos_pos": 0.9634201526641846,
+    "cos_neg": 0.9624224305152893,
+    "collapse": -0.9472368955612183,
+    "edrift_p": 0.07989571988582611,
+    "edrift_n": 0.06900421530008316
+  },
+  "target_replace": [
+    "Qwen3Attention"
+  ],
+  "kind": "language_model",
+  "prefix": "lora_te",
+  "delimiter": "-",
+  "train_method": "full",
+  "unit_scale": 1.0,
+  "endreg": {
+    "enabled": true,
+    "weight": 1.0,
+    "frames": 250,
+    "seed": 7,
+    "rows_ended_naturally": 1,
+    "final_edrift_p": 0.060291916355490686,
+    "final_edrift_n": 0.06002865001559257
+  },
+  "plus_label": "Female",
+  "minus_label": "Male",
+  "recommended_range": [
+    -2.0,
+    2.0
+  ],
+  "prompts_file": "conceptmod/textsliders/data/prompts-gender-v4.yaml",
+  "early_stop": {
+    "enabled": true,
+    "fired": false,
+    "window": 50,
+    "min_steps": 100,
+    "cos": 0.97,
+    "collapse": -0.95,
+    "perc": 0.2,
+    "metrics": {
+      "cos_pos": 0.9720010232925415,
+      "cos_neg": 0.9719867610931396,
+      "collapse": -0.954462149143219,
+      "pperc": 0.2344539639353752,
+      "nperc": 0.23543490827083588
+    }
+  }
+}

weights/gender-lm-v4/gender-lm-v4_last.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b23e3200e1fba4efb54499ba93d22d5c1a7c15261923deb19c20e453c022e33f
+size 30723552

weights/live-lm-v5/live-lm-v5_last.json

@@ -0,0 +1,79 @@
+{
+  "schema": 3,
+  "name": "live-lm-v5",
+  "checkpoint": "/ml2/music/sliders-conceptmod/models/live-lm-v5/live-lm-v5_last.safetensors",
+  "weights": "/ml2/music/sliders-conceptmod/models/live-lm-v5/live-lm-v5_last.safetensors",
+  "modules": 144,
+  "rank": 8,
+  "alpha": 8.0,
+  "steps": 800,
+  "steps_budget": 800,
+  "lr": 0.0005,
+  "rows": 6,
+  "symmetric": true,
+  "common_beta": 0.0,
+  "first": {
+    "step": 1,
+    "row": 0,
+    "loss": 0.04443066567182541,
+    "pperc": 1.0064728260040283,
+    "nperc": 1.001420497894287,
+    "cos_pos": -0.02850652113556862,
+    "cos_neg": 0.02850652113556862,
+    "collapse": 1.0,
+    "edrift_p": 0.0,
+    "edrift_n": 0.0
+  },
+  "last": {
+    "step": 800,
+    "row": 1,
+    "loss": 0.012794841080904007,
+    "pperc": 0.34088581800460815,
+    "nperc": 0.37231141328811646,
+    "cos_pos": 0.940139889717102,
+    "cos_neg": 0.9281188249588013,
+    "collapse": -0.955425500869751,
+    "edrift_p": 0.0671699270606041,
+    "edrift_n": 0.06693821400403976
+  },
+  "target_replace": [
+    "Qwen3Attention"
+  ],
+  "kind": "language_model",
+  "prefix": "lora_te",
+  "delimiter": "-",
+  "train_method": "full",
+  "unit_scale": 1.0,
+  "endreg": {
+    "enabled": true,
+    "weight": 1.0,
+    "frames": 250,
+    "seed": 7,
+    "rows_ended_naturally": 0,
+    "final_edrift_p": 0.06764324523508548,
+    "final_edrift_n": 0.06896134436130524
+  },
+  "plus_label": "Live",
+  "minus_label": "Studio",
+  "recommended_range": [
+    -2.0,
+    2.0
+  ],
+  "prompts_file": "conceptmod/textsliders/data/prompts-live-v4.yaml",
+  "early_stop": {
+    "enabled": true,
+    "fired": false,
+    "window": 50,
+    "min_steps": 100,
+    "cos": 0.97,
+    "collapse": -0.95,
+    "perc": 0.2,
+    "metrics": {
+      "cos_pos": 0.9058725821971894,
+      "cos_neg": 0.8977998125553132,
+      "collapse": -0.9137211871147156,
+      "pperc": 0.4195082801580429,
+      "nperc": 0.4369349026679993
+    }
+  }
+}

weights/live-lm-v5/live-lm-v5_last.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:348341c4f3c1008e0e425fa49b0e4a928a337c8c7c0a6d8f8b1afcf09696717e
+size 30723552

weights/rapslow-lm-v4/rapslow-lm-v4_last.json

@@ -0,0 +1,79 @@
+{
+  "schema": 3,
+  "name": "rapslow-lm-v4",
+  "checkpoint": "/ml2/music/sliders-conceptmod/models/rapslow-lm-v4/rapslow-lm-v4_last.safetensors",
+  "weights": "/ml2/music/sliders-conceptmod/models/rapslow-lm-v4/rapslow-lm-v4_last.safetensors",
+  "modules": 144,
+  "rank": 8,
+  "alpha": 8.0,
+  "steps": 800,
+  "steps_budget": 800,
+  "lr": 0.0005,
+  "rows": 6,
+  "symmetric": true,
+  "common_beta": 0.0,
+  "first": {
+    "step": 1,
+    "row": 0,
+    "loss": 0.04510205611586571,
+    "pperc": 0.9948540925979614,
+    "nperc": 1.0180939435958862,
+    "cos_pos": 0.10208363085985184,
+    "cos_neg": -0.10208363085985184,
+    "collapse": 0.9999999403953552,
+    "edrift_p": 0.0,
+    "edrift_n": 0.0
+  },
+  "last": {
+    "step": 800,
+    "row": 1,
+    "loss": 0.0125714847818017,
+    "pperc": 0.3322715759277344,
+    "nperc": 0.3056045174598694,
+    "cos_pos": 0.9432080388069153,
+    "cos_neg": 0.9521855115890503,
+    "collapse": -0.9507349729537964,
+    "edrift_p": 0.06929578632116318,
+    "edrift_n": 0.07001179456710815
+  },
+  "target_replace": [
+    "Qwen3Attention"
+  ],
+  "kind": "language_model",
+  "prefix": "lora_te",
+  "delimiter": "-",
+  "train_method": "full",
+  "unit_scale": 1.0,
+  "endreg": {
+    "enabled": true,
+    "weight": 1.0,
+    "frames": 250,
+    "seed": 7,
+    "rows_ended_naturally": 0,
+    "final_edrift_p": 0.08115319438278674,
+    "final_edrift_n": 0.07886697210371495
+  },
+  "plus_label": "Rap",
+  "minus_label": "Slow",
+  "recommended_range": [
+    -2.0,
+    2.0
+  ],
+  "prompts_file": "conceptmod/textsliders/data/prompts-rapslow-v4.yaml",
+  "early_stop": {
+    "enabled": true,
+    "fired": false,
+    "window": 50,
+    "min_steps": 100,
+    "cos": 0.97,
+    "collapse": -0.95,
+    "perc": 0.2,
+    "metrics": {
+      "cos_pos": 0.958186708688736,
+      "cos_neg": 0.9544536030292511,
+      "collapse": -0.9589180564880371,
+      "pperc": 0.2847372156381607,
+      "nperc": 0.2979007542133331
+    }
+  }
+}

weights/rapslow-lm-v4/rapslow-lm-v4_last.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:413c13d5e580232ce1ce228f6d868e38db48d66f5aa40018ef3ccbce493d790b
+size 30723552

weights/rhyme-lm-v5/rhyme-lm-v5_last.json

@@ -0,0 +1,79 @@
+{
+  "schema": 3,
+  "name": "rhyme-lm-v5",
+  "checkpoint": "/ml2/music/sliders-conceptmod/models/rhyme-lm-v5/rhyme-lm-v5_last.safetensors",
+  "weights": "/ml2/music/sliders-conceptmod/models/rhyme-lm-v5/rhyme-lm-v5_last.safetensors",
+  "modules": 144,
+  "rank": 8,
+  "alpha": 8.0,
+  "steps": 800,
+  "steps_budget": 800,
+  "lr": 0.0005,
+  "rows": 6,
+  "symmetric": true,
+  "common_beta": 0.0,
+  "first": {
+    "step": 1,
+    "row": 0,
+    "loss": 0.03842737898230553,
+    "pperc": 0.9976922869682312,
+    "nperc": 1.0140079259872437,
+    "cos_pos": 0.07553429901599884,
+    "cos_neg": -0.07553429901599884,
+    "collapse": 1.0,
+    "edrift_p": 0.0,
+    "edrift_n": 0.0
+  },
+  "last": {
+    "step": 800,
+    "row": 1,
+    "loss": 0.011187395080924034,
+    "pperc": 0.29871201515197754,
+    "nperc": 0.31096577644348145,
+    "cos_pos": 0.954346239566803,
+    "cos_neg": 0.9505065679550171,
+    "collapse": -0.9642457962036133,
+    "edrift_p": 0.06575663387775421,
+    "edrift_n": 0.06813128292560577
+  },
+  "target_replace": [
+    "Qwen3Attention"
+  ],
+  "kind": "language_model",
+  "prefix": "lora_te",
+  "delimiter": "-",
+  "train_method": "full",
+  "unit_scale": 1.0,
+  "endreg": {
+    "enabled": true,
+    "weight": 1.0,
+    "frames": 250,
+    "seed": 7,
+    "rows_ended_naturally": 0,
+    "final_edrift_p": 0.07013336338102817,
+    "final_edrift_n": 0.06966199301183224
+  },
+  "plus_label": "Rhyme",
+  "minus_label": "Prose",
+  "recommended_range": [
+    -2.0,
+    2.0
+  ],
+  "prompts_file": "conceptmod/textsliders/data/prompts-rhyme-v4.yaml",
+  "early_stop": {
+    "enabled": false,
+    "fired": false,
+    "window": 50,
+    "min_steps": 100,
+    "cos": 0.97,
+    "collapse": -0.95,
+    "perc": 0.2,
+    "metrics": {
+      "cos_pos": 0.9360902714729309,
+      "cos_neg": 0.9468878376483917,
+      "collapse": -0.9473911559581757,
+      "pperc": 0.3495775777101517,
+      "nperc": 0.3220868802070618
+    }
+  }
+}

weights/rhyme-lm-v5/rhyme-lm-v5_last.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd24e89d8a5ee6810c2a827814ce7424f3f1aca9650dce16790075880b5ff9e5
+size 30723552

weights/tempo-lm-v4/tempo-lm-v4_last.json

@@ -0,0 +1,79 @@
+{
+  "schema": 3,
+  "name": "tempo-lm-v4",
+  "checkpoint": "/ml2/music/sliders-conceptmod/models/tempo-lm-v4/tempo-lm-v4_last.safetensors",
+  "weights": "/ml2/music/sliders-conceptmod/models/tempo-lm-v4/tempo-lm-v4_last.safetensors",
+  "modules": 144,
+  "rank": 8,
+  "alpha": 8.0,
+  "steps": 800,
+  "steps_budget": 800,
+  "lr": 0.0005,
+  "rows": 6,
+  "symmetric": true,
+  "common_beta": 0.0,
+  "first": {
+    "step": 1,
+    "row": 0,
+    "loss": 0.05744679272174835,
+    "pperc": 0.9987309575080872,
+    "nperc": 1.0085179805755615,
+    "cos_pos": 0.057536084204912186,
+    "cos_neg": -0.057536084204912186,
+    "collapse": 1.0,
+    "edrift_p": 0.0,
+    "edrift_n": 0.0
+  },
+  "last": {
+    "step": 800,
+    "row": 1,
+    "loss": 0.014287056401371956,
+    "pperc": 0.2998714745044708,
+    "nperc": 0.25132104754447937,
+    "cos_pos": 0.9545653462409973,
+    "cos_neg": 0.9681466817855835,
+    "collapse": -0.9599778652191162,
+    "edrift_p": 0.06434958428144455,
+    "edrift_n": 0.06956690549850464
+  },
+  "target_replace": [
+    "Qwen3Attention"
+  ],
+  "kind": "language_model",
+  "prefix": "lora_te",
+  "delimiter": "-",
+  "train_method": "full",
+  "unit_scale": 1.0,
+  "endreg": {
+    "enabled": true,
+    "weight": 1.0,
+    "frames": 250,
+    "seed": 7,
+    "rows_ended_naturally": 0,
+    "final_edrift_p": 0.07175659276545047,
+    "final_edrift_n": 0.07057853572070599
+  },
+  "plus_label": "Fast",
+  "minus_label": "Slow",
+  "recommended_range": [
+    -2.0,
+    2.0
+  ],
+  "prompts_file": "conceptmod/textsliders/data/prompts-tempo-v4.yaml",
+  "early_stop": {
+    "enabled": true,
+    "fired": false,
+    "window": 50,
+    "min_steps": 100,
+    "cos": 0.97,
+    "collapse": -0.95,
+    "perc": 0.2,
+    "metrics": {
+      "cos_pos": 0.9520531296730042,
+      "cos_neg": 0.9558060193061828,
+      "collapse": -0.9517069292068482,
+      "pperc": 0.3006351378560066,
+      "nperc": 0.29288982510566713
+    }
+  }
+}

weights/tempo-lm-v4/tempo-lm-v4_last.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf29de200d3d25905daa7b80d01c67caba05f60987cdefc5b6dbe9a7c23d8abb
+size 30723552

weights/triphop-lm-v4/triphop-lm-v4_last.json

@@ -0,0 +1,79 @@
+{
+  "schema": 3,
+  "name": "triphop-lm-v4",
+  "checkpoint": "/ml2/music/sliders-conceptmod/models/triphop-lm-v4/triphop-lm-v4_last.safetensors",
+  "weights": "/ml2/music/sliders-conceptmod/models/triphop-lm-v4/triphop-lm-v4_last.safetensors",
+  "modules": 144,
+  "rank": 8,
+  "alpha": 8.0,
+  "steps": 800,
+  "steps_budget": 800,
+  "lr": 0.0005,
+  "rows": 4,
+  "symmetric": true,
+  "common_beta": 0.0,
+  "first": {
+    "step": 1,
+    "row": 0,
+    "loss": 0.0205980371683836,
+    "pperc": 1.0161914825439453,
+    "nperc": 0.9956115484237671,
+    "cos_pos": -0.09471139311790466,
+    "cos_neg": 0.09471139311790466,
+    "collapse": 1.0,
+    "edrift_p": 0.0,
+    "edrift_n": 0.0
+  },
+  "last": {
+    "step": 800,
+    "row": 3,
+    "loss": 0.007938449271023273,
+    "pperc": 0.23580722510814667,
+    "nperc": 0.21540263295173645,
+    "cos_pos": 0.9728948473930359,
+    "cos_neg": 0.9770852327346802,
+    "collapse": -0.9601766467094421,
+    "edrift_p": 0.05283118411898613,
+    "edrift_n": 0.06300079077482224
+  },
+  "target_replace": [
+    "Qwen3Attention"
+  ],
+  "kind": "language_model",
+  "prefix": "lora_te",
+  "delimiter": "-",
+  "train_method": "full",
+  "unit_scale": 1.0,
+  "endreg": {
+    "enabled": true,
+    "weight": 1.0,
+    "frames": 250,
+    "seed": 7,
+    "rows_ended_naturally": 0,
+    "final_edrift_p": 0.05935625784099102,
+    "final_edrift_n": 0.05895587995648384
+  },
+  "plus_label": "Trip-hop",
+  "minus_label": "Pop",
+  "recommended_range": [
+    -2.0,
+    2.0
+  ],
+  "prompts_file": "conceptmod/textsliders/data/prompts-triphop-v4.yaml",
+  "early_stop": {
+    "enabled": true,
+    "fired": false,
+    "window": 50,
+    "min_steps": 100,
+    "cos": 0.97,
+    "collapse": -0.95,
+    "perc": 0.2,
+    "metrics": {
+      "cos_pos": 0.9662414407730102,
+      "cos_neg": 0.9686378180980683,
+      "collapse": -0.9532504975795746,
+      "pperc": 0.2590042370557785,
+      "nperc": 0.24962790429592133
+    }
+  }
+}

weights/triphop-lm-v4/triphop-lm-v4_last.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d305fd8cad7f966063ebe806a27baf8c3b6b7062bbacf4da5e563d80a79af44a
+size 30723552