@tuned-tensor/cli 0.10.0 → 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.
- package/README.md +21 -6
- package/dist/{chunk-DAP2Z7EZ.js → chunk-HDGBDHOB.js} +474 -74
- package/dist/chunk-HDGBDHOB.js.map +1 -0
- package/dist/index.js +235 -84
- package/dist/index.js.map +1 -1
- package/dist/local-runtime.js +52 -19
- package/dist/local-runtime.js.map +1 -1
- package/docs/local-runtime/architecture.md +8 -2
- package/docs/local-runtime/spark.md +39 -2
- package/package.json +5 -1
- package/training/local-runner/pyproject.toml +2 -2
- package/training/local-runner/src/evaluate.py +44 -18
- package/training/local-runner/src/model_contract.py +134 -14
- package/training/local-runner/src/prefetch.py +25 -8
- package/training/local-runner/src/serve.py +14 -6
- package/training/local-runner/src/sft_data.py +7 -3
- package/training/local-runner/src/train.py +48 -14
- package/training/local-runner/uv.lock +54 -54
- package/dist/chunk-DAP2Z7EZ.js.map +0 -1
package/README.md
CHANGED
|
@@ -63,6 +63,11 @@ Selection metadata can be overridden per process with
|
|
|
63
63
|
After configuration, run `tt` to open one terminal for conversation and
|
|
64
64
|
commands:
|
|
65
65
|
|
|
66
|
+
Before the shell opens, `tt` performs a short, non-blocking npm version check.
|
|
67
|
+
If a newer stable release is available it recommends
|
|
68
|
+
`npm install -g @tuned-tensor/cli@latest`; offline or unavailable registry
|
|
69
|
+
checks are ignored.
|
|
70
|
+
|
|
66
71
|
```text
|
|
67
72
|
tt cloud support-agent › What happened in my latest training run?
|
|
68
73
|
TT Your latest run completed...
|
|
@@ -114,9 +119,13 @@ ID. If a response or final state write is lost after dispatch, the action is
|
|
|
114
119
|
retained as `outcome_unknown`, cannot be retried automatically, and directs the
|
|
115
120
|
user to inspect the remote resource. `/reject` never calls a mutation endpoint.
|
|
116
121
|
|
|
117
|
-
Useful shell controls include `/help`, `/status`, `/context`, `/
|
|
118
|
-
`/model`, `/cd`, `/clear`, and `/exit`. `/
|
|
119
|
-
|
|
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
|
|
120
129
|
`/model <id>` activates a verified local serving model. Configure the assistant
|
|
121
130
|
model explicitly with `tt agent configure`. The shell keeps normal terminal
|
|
122
131
|
scrollback and command history only for the current process.
|
|
@@ -216,9 +225,15 @@ tt local models verify local-<run-id>
|
|
|
216
225
|
tt local models serve local-<run-id> --config local-runner.json
|
|
217
226
|
```
|
|
218
227
|
|
|
219
|
-
Local currently certifies text SFT with `Qwen/Qwen3.5-2B`,
|
|
220
|
-
|
|
221
|
-
|
|
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.
|
|
222
237
|
|
|
223
238
|
Activation is optional and requires the run to pass a configured
|
|
224
239
|
`generalRegression` gate. Once activated, use
|