@tuned-tensor/cli 0.11.0 → 0.12.0-beta.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (19) hide show
  1. package/README.md +16 -6
  2. package/dist/{chunk-DAP2Z7EZ.js → chunk-HDGBDHOB.js} +474 -74
  3. package/dist/chunk-HDGBDHOB.js.map +1 -0
  4. package/dist/index.js +135 -83
  5. package/dist/index.js.map +1 -1
  6. package/dist/local-runtime.js +52 -19
  7. package/dist/local-runtime.js.map +1 -1
  8. package/docs/local-runtime/architecture.md +8 -2
  9. package/docs/local-runtime/spark.md +39 -2
  10. package/package.json +5 -1
  11. package/training/local-runner/pyproject.toml +2 -2
  12. package/training/local-runner/src/evaluate.py +44 -18
  13. package/training/local-runner/src/model_contract.py +134 -14
  14. package/training/local-runner/src/prefetch.py +25 -8
  15. package/training/local-runner/src/serve.py +14 -6
  16. package/training/local-runner/src/sft_data.py +7 -3
  17. package/training/local-runner/src/train.py +48 -14
  18. package/training/local-runner/uv.lock +54 -54
  19. package/dist/chunk-DAP2Z7EZ.js.map +0 -1
package/README.md CHANGED
@@ -119,9 +119,13 @@ ID. If a response or final state write is lost after dispatch, the action is
119
119
  retained as `outcome_unknown`, cannot be retried automatically, and directs the
120
120
  user to inspect the remote resource. `/reject` never calls a mutation endpoint.
121
121
 
122
- Useful shell controls include `/help`, `/status`, `/context`, `/mode`,
123
- `/model`, `/cd`, `/clear`, and `/exit`. `/model` retains its workflow meaning:
124
- it shows the active local serving model or cloud spec base model, and
122
+ Useful shell controls include `/help`, `/status`, `/context`, `/cloud`,
123
+ `/local`, `/mode`, `/model`, `/cd`, `/clear`, and `/exit`. `/cloud` and
124
+ `/local` switch the workflow in one word (aliases for `/mode cloud|local`).
125
+ The shell banner and `/context` surface the assistant's configured provider and
126
+ model (`agent provider/model`) separately from the workflow model, so the
127
+ model answering your prompts is always visible. `/model` retains its workflow
128
+ meaning: it shows the active local serving model or cloud spec base model, and
125
129
  `/model <id>` activates a verified local serving model. Configure the assistant
126
130
  model explicitly with `tt agent configure`. The shell keeps normal terminal
127
131
  scrollback and command history only for the current process.
@@ -221,9 +225,15 @@ tt local models verify local-<run-id>
221
225
  tt local models serve local-<run-id> --config local-runner.json
222
226
  ```
223
227
 
224
- Local currently certifies text SFT with `Qwen/Qwen3.5-2B`, LoRA/PEFT, and CUDA.
225
- Evaluation and serving may use CPU. Training artifacts, datasets, and model
226
- weights remain on the execution host.
228
+ Local currently certifies text SFT with `Qwen/Qwen3.5-2B`,
229
+ `nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16`, and
230
+ `meta-models/Muse-Glimmer-30B`, using LoRA/PEFT and CUDA. The 30B/3B-active
231
+ Nemotron path is intended for DGX Spark-class unified memory, uses activation
232
+ checkpointing, and adapts bounded shared attention/Mamba projections rather
233
+ than every routed expert matrix. Muse Glimmer is a vision-language checkpoint
234
+ whose text tower is fine-tuned through the image-text-to-text loader with the
235
+ vision tower frozen and unused. Evaluation and serving may use CPU. Training
236
+ artifacts, datasets, and model weights remain on the execution host.
227
237
 
228
238
  Activation is optional and requires the run to pass a configured
229
239
  `generalRegression` gate. Once activated, use