@tuned-tensor/cli 0.13.1-beta.2 → 0.13.1-beta.4
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 +20 -2
- package/dist/{chunk-2P2RL23O.js → chunk-7VIUHQBI.js} +994 -845
- package/dist/chunk-7VIUHQBI.js.map +1 -0
- package/dist/index.js +257 -49
- package/dist/index.js.map +1 -1
- package/dist/local-runtime.js +32 -9
- package/dist/local-runtime.js.map +1 -1
- package/docs/local-runtime/architecture.md +4 -4
- package/examples/local-runtime/dry-runner.json +2 -2
- package/examples/local-runtime/local-runner.json +2 -2
- package/package.json +5 -5
- package/dist/chunk-2P2RL23O.js.map +0 -1
- /package/training/{local-runner → adapter}/pyproject.toml +0 -0
- /package/training/{local-runner → adapter}/src/evaluate.py +0 -0
- /package/training/{local-runner → adapter}/src/model_contract.py +0 -0
- /package/training/{local-runner → adapter}/src/prefetch.py +0 -0
- /package/training/{local-runner → adapter}/src/serve.py +0 -0
- /package/training/{local-runner → adapter}/src/sft_data.py +0 -0
- /package/training/{local-runner → adapter}/src/train.py +0 -0
- /package/training/{local-runner → adapter}/uv.lock +0 -0
package/README.md
CHANGED
|
@@ -130,9 +130,27 @@ roots, existing targets, and unsupported spec fields. Known CLI commands such
|
|
|
130
130
|
as `runs list`, `doctor`, and `models list` still execute directly. Prefix a
|
|
131
131
|
command with `:` when you want to make that intent explicit.
|
|
132
132
|
|
|
133
|
+
The agent can describe the built-in adapter and foundation pipeline stages,
|
|
134
|
+
including optional foundation RL, and returns the exact `tt pipeline` commands
|
|
135
|
+
for initialization, validation, planning, dry-run, and execution. It can also
|
|
136
|
+
search public Hugging Face model or dataset metadata for foundation and
|
|
137
|
+
fine-tuning discovery. A Hub search sends only the search query to
|
|
138
|
+
`huggingface.co`; it uses no Hugging Face token, downloads nothing, and omits
|
|
139
|
+
model-card and dataset-card text. Search results are discovery metadata, not a
|
|
140
|
+
guarantee that a model or dataset is compatible with a TT workflow.
|
|
141
|
+
|
|
142
|
+
For educational questions about how or why training works, the agent can read
|
|
143
|
+
the exact Python implementation shipped with the running TT build. The source
|
|
144
|
+
tool is limited to named foundation components (tokenizer, pretraining, model,
|
|
145
|
+
data, SFT, RL, evaluation, and common helpers) and adapter components (training,
|
|
146
|
+
data, model contract, and evaluation); it cannot read arbitrary paths. Answers
|
|
147
|
+
separate behavior directly visible in the code from inferred rationale and
|
|
148
|
+
include the inspected package-relative path and source hash.
|
|
149
|
+
|
|
133
150
|
Agent conversations are durable and local under
|
|
134
|
-
`~/.
|
|
135
|
-
user-only directory/file permissions.
|
|
151
|
+
`~/.tuned-tensor/agent/threads` (or `$TUNED_TENSOR_HOME/agent/threads`), with
|
|
152
|
+
user-only directory/file permissions. Existing XDG agent state remains in use
|
|
153
|
+
until the new agent directory exists. Persistence normally retains up to 100
|
|
136
154
|
recent threads; unsettled safety records are never pruned even when that exceeds
|
|
137
155
|
the cap. Each thread is limited to 1 MB, 200 messages, and 200 actions; one-way
|
|
138
156
|
action claims are capped at 1,000. Use `/new` to start one, `/threads` to
|