annotuneagent 0.1.0__tar.gz → 0.1.1__tar.gz

{annotuneagent-0.1.0 → annotuneagent-0.1.1}/CHANGELOG.md

@@ -4,10 +4,26 @@ All notable changes to this project are documented in this file.
 
 ## [Unreleased]
 
-### Added
+## [0.1.1] - 2026-11-09
+
+### Fixed
+
+- A running server is checked with `/props` and `/lora-adapters` when `/proc`
+  is unavailable, so Windows and macOS can reuse a server this package did
+  not start.
+- A `TaskEndpoint` URL calls `/completion` and enables one LoRA adapter for
+  that request. When `maxlen` is omitted, name allows 2048 tokens and ML
+  allows 8192. An ML prompt does not fit in the name limit.
+- A model request waits 600 seconds. `ANNOTUNE_TIMEOUT` or the `timeout`
+  argument overrides it. Health checks stay at 2 seconds.
+- On Windows, a process check does not call `os.kill`.
+
+### Changed
 
-- PyPI publish workflow. `pip install annotuneagent` after the package is
-  published. Publishing is manual from GitHub Actions.
+- The load-model tutorial points Windows users to
+  [AnnoTuneLoader](https://github.com/leelieber2025/AnnoTuneLoader).
+- A tutorial explains each field of the `run` report. The `Queries` paragraph
+  is text to paste into another model.
 
 ## [0.1.0] - 2026-11-09
 

{annotuneagent-0.1.0 → annotuneagent-0.1.1}/CITATION.cff

@@ -17,7 +17,7 @@ authors:
     orcid: "https://orcid.org/0000-0002-9276-9129"
 repository-code: "https://github.com/leelieber2025/AnnoTuneAgent"
 license: Apache-2.0
-version: 0.1.0
+version: 0.1.1
 date-released: "2026-11-09"
 keywords:
   - single-cell

{annotuneagent-0.1.0 → annotuneagent-0.1.1}/PKG-INFO

@@ -1,15 +1,16 @@
 Metadata-Version: 2.4
 Name: annotuneagent
-Version: 0.1.0
+Version: 0.1.1
 Summary: Cluster-level cell-type annotation from marker, name, and ML results
 Author-email: "Zhao Li (李钊)" <leelieber@gmail.com>
 License-Expression: Apache-2.0
 Project-URL: Homepage, https://github.com/leelieber2025/AnnoTuneAgent
-Project-URL: Documentation, https://app.readthedocs.org/projects/annotuneagent/
+Project-URL: Documentation, https://annotuneagent.readthedocs.io/
 Project-URL: Changelog, https://github.com/leelieber2025/AnnoTuneAgent/blob/main/CHANGELOG.md
 Project-URL: Source, https://github.com/leelieber2025/AnnoTuneAgent
 Project-URL: Issues, https://github.com/leelieber2025/AnnoTuneAgent/issues
 Project-URL: PyPI, https://pypi.org/project/annotuneagent/
+Project-URL: Zenodo, https://doi.org/10.5281/zenodo.24092828
 Keywords: single-cell,RNA-seq,cell type annotation,marker genes,bioinformatics
 Classifier: Development Status :: 3 - Alpha
 Classifier: Intended Audience :: Science/Research
@@ -52,11 +53,15 @@ Dynamic: license-file
 
 # AnnoTuneAgent
 
+[![PyPI version](https://img.shields.io/pypi/v/annotuneagent.svg)](https://pypi.org/project/annotuneagent/)
+[![PyPI downloads](https://img.shields.io/pepy/dt/annotuneagent.svg?label=PyPI%20downloads)](https://pepy.tech/project/annotuneagent)
+[![Python versions](https://img.shields.io/pypi/pyversions/annotuneagent.svg)](https://pypi.org/project/annotuneagent/)
+[![Documentation Status](https://readthedocs.org/projects/annotuneagent/badge/?version=latest)](https://annotuneagent.readthedocs.io/en/latest/?badge=latest)
+[![CI](https://github.com/leelieber2025/AnnoTuneAgent/actions/workflows/tests.yml/badge.svg)](https://github.com/leelieber2025/AnnoTuneAgent/actions/workflows/tests.yml)
 [![License](https://img.shields.io/badge/license-Apache--2.0-blue.svg)](LICENSE)
-[![PyPI](https://img.shields.io/pypi/v/annotuneagent.svg)](https://pypi.org/project/annotuneagent/)
-[![Documentation](https://app.readthedocs.org/projects/annotuneagent/badge/?version=latest)](https://app.readthedocs.org/projects/annotuneagent/)
+[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.24092828.svg)](https://doi.org/10.5281/zenodo.24092828)
 
-Documentation: [Read the Docs](https://app.readthedocs.org/projects/annotuneagent/).
+Documentation: [Read the Docs](https://annotuneagent.readthedocs.io/en/latest/).
 
 **Cluster-level cell-type annotation for single-cell RNA-seq, with two
 fine-tuned language models that run locally.**
@@ -127,7 +132,13 @@ Harmony, and Scanorama. If you select one that is missing, `prepare` prints
 the install command and leaves the matrix unchanged. See
 [Installation](docs/installation.md).
 
-From a checkout: `pip install -e '.[dev]'`.
+From a checkout:
+
+```bash
+git clone https://github.com/leelieber2025/AnnoTuneAgent.git
+cd AnnoTuneAgent
+pip install -e ".[dev]"
+```
 
 ## First run
 
@@ -196,7 +207,12 @@ The tokenizer is installed with the package. See
 
 ## Status
 
-**0.1.0 (Alpha).** Import as `import annotuneagent as ata`.
+**0.1.1 (Alpha).** Import as `import annotuneagent as ata`.
+
+## Citation
+
+For the software record, cite the [Zenodo DOI](https://doi.org/10.5281/zenodo.24092828).
+Pin the package version used in the analysis, for example `annotuneagent==0.1.1`.
 
 ## License
 

{annotuneagent-0.1.0 → annotuneagent-0.1.1}/README.md

@@ -1,10 +1,14 @@
 # AnnoTuneAgent
 
+[![PyPI version](https://img.shields.io/pypi/v/annotuneagent.svg)](https://pypi.org/project/annotuneagent/)
+[![PyPI downloads](https://img.shields.io/pepy/dt/annotuneagent.svg?label=PyPI%20downloads)](https://pepy.tech/project/annotuneagent)
+[![Python versions](https://img.shields.io/pypi/pyversions/annotuneagent.svg)](https://pypi.org/project/annotuneagent/)
+[![Documentation Status](https://readthedocs.org/projects/annotuneagent/badge/?version=latest)](https://annotuneagent.readthedocs.io/en/latest/?badge=latest)
+[![CI](https://github.com/leelieber2025/AnnoTuneAgent/actions/workflows/tests.yml/badge.svg)](https://github.com/leelieber2025/AnnoTuneAgent/actions/workflows/tests.yml)
 [![License](https://img.shields.io/badge/license-Apache--2.0-blue.svg)](LICENSE)
-[![PyPI](https://img.shields.io/pypi/v/annotuneagent.svg)](https://pypi.org/project/annotuneagent/)
-[![Documentation](https://app.readthedocs.org/projects/annotuneagent/badge/?version=latest)](https://app.readthedocs.org/projects/annotuneagent/)
+[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.24092828.svg)](https://doi.org/10.5281/zenodo.24092828)
 
-Documentation: [Read the Docs](https://app.readthedocs.org/projects/annotuneagent/).
+Documentation: [Read the Docs](https://annotuneagent.readthedocs.io/en/latest/).
 
 **Cluster-level cell-type annotation for single-cell RNA-seq, with two
 fine-tuned language models that run locally.**
@@ -75,7 +79,13 @@ Harmony, and Scanorama. If you select one that is missing, `prepare` prints
 the install command and leaves the matrix unchanged. See
 [Installation](docs/installation.md).
 
-From a checkout: `pip install -e '.[dev]'`.
+From a checkout:
+
+```bash
+git clone https://github.com/leelieber2025/AnnoTuneAgent.git
+cd AnnoTuneAgent
+pip install -e ".[dev]"
+```
 
 ## First run
 
@@ -144,7 +154,12 @@ The tokenizer is installed with the package. See
 
 ## Status
 
-**0.1.0 (Alpha).** Import as `import annotuneagent as ata`.
+**0.1.1 (Alpha).** Import as `import annotuneagent as ata`.
+
+## Citation
+
+For the software record, cite the [Zenodo DOI](https://doi.org/10.5281/zenodo.24092828).
+Pin the package version used in the analysis, for example `annotuneagent==0.1.1`.
 
 ## License
 

{annotuneagent-0.1.0 → annotuneagent-0.1.1}/docs/index.md

@@ -1,5 +1,12 @@
 # AnnoTuneAgent Documentation
 
+[![PyPI version](https://img.shields.io/pypi/v/annotuneagent.svg)](https://pypi.org/project/annotuneagent/)
+[![PyPI downloads](https://img.shields.io/pepy/dt/annotuneagent.svg?label=PyPI%20downloads)](https://pepy.tech/project/annotuneagent)
+[![Python versions](https://img.shields.io/pypi/pyversions/annotuneagent.svg)](https://pypi.org/project/annotuneagent/)
+[![CI](https://github.com/leelieber2025/AnnoTuneAgent/actions/workflows/tests.yml/badge.svg)](https://github.com/leelieber2025/AnnoTuneAgent/actions/workflows/tests.yml)
+[![License](https://img.shields.io/badge/license-Apache--2.0-blue.svg)](https://github.com/leelieber2025/AnnoTuneAgent/blob/main/LICENSE)
+[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.24092828.svg)](https://doi.org/10.5281/zenodo.24092828)
+
 AnnoTuneAgent annotates clustered scRNA-seq. `prepare` runs QC, gene
 selection, and Leiden clustering. `run` collects three results for each
 cluster: a marker score, a name-model result, and an ML-model result. A
@@ -52,6 +59,7 @@ scFair supplies the gene list and the resolution is half of its suggestion.
 | {doc}`installation`, {doc}`quickstart` | Install, then a first call |
 | {doc}`tutorials/pbmc8k_annotation` | PBMC 8k: `prepare`, then the name and ML models |
 | {doc}`tutorials/load_model` | Download the weights and run a local or remote model server |
+| {doc}`tutorials/report` | What each part of the `run` report is for, including the text to paste |
 | {doc}`api/index` | Function signatures |
 | {doc}`marker_sources` | Marker reference sources and verification decisions |
 

{annotuneagent-0.1.0 → annotuneagent-0.1.1}/docs/installation.md

@@ -6,14 +6,21 @@ Python 3.10–3.13.
 pip install annotuneagent
 ```
 
-From a checkout, `pip install -e .` installs this tree instead.
-
 ```python
 import annotuneagent as ata
 print(ata.__version__)
 ```
 
-The version is set in `src/annotuneagent/_version.py` (currently **0.1.0**).
+The version is set in `src/annotuneagent/_version.py` (currently **0.1.1**).
+
+From a checkout:
+
+```bash
+git clone https://github.com/leelieber2025/AnnoTuneAgent.git
+cd AnnoTuneAgent
+pip install -e ".[dev]"
+pip install -e ".[docs]"
+```
 
 ## Dependencies
 
@@ -32,13 +39,6 @@ package. These are optional:
 If you select an optional method and its package is missing, `prepare` prints
 the install command and leaves the matrix unchanged. ComBat is part of Scanpy.
 
-Other extras:
-
-```bash
-pip install -e ".[docs]"    # build this documentation from a checkout
-pip install -e ".[dev]"     # pytest and ruff
-```
-
 ## Model weights
 
 The GGUF weights are not in the Python package. Download them from the

{annotuneagent-0.1.0 → annotuneagent-0.1.1}/docs/quickstart.md

@@ -184,6 +184,10 @@ panel, not to each gene separately.
   and UMAP and dotplot calls. Edit the order or labels and rerun the snippet to
   regenerate the plots.
 
+{doc}`tutorials/report` explains every report field. The printed `Queries`
+block is text to paste into another model. The heading is the local note.
+The paragraph under it is the question.
+
 A Cell Ontology id is the exact map row for the written label. An unreviewed id
 is marked pending.
 

{annotuneagent-0.1.0 → annotuneagent-0.1.1}/docs/tutorials/index.md

@@ -5,10 +5,11 @@ figures and the UMAP. `run` collects the marker score, the name-model result,
 and the ML-model result, and writes the title. The model server stays loaded
 between calls.
 
-| Notebook | Contents |
+| Page | Contents |
 |----------|----------|
 | {doc}`pbmc8k_annotation` | Public PBMC 8k: `prepare`, then the name and ML models |
 | {doc}`load_model` | Download the weights, then use a local server or connect to a remote one |
+| {doc}`report` | What each part of the `run` report is for, including the text to paste |
 
 {doc}`pbmc8k_annotation` uses the public 10x PBMC 8k count matrix (7,750 cells),
 not the Scanpy PBMC 3k set. The matrix contains cells only, so AmbiDose is off.
@@ -42,6 +43,15 @@ Run a local server at `http://127.0.0.1:8080`, or connect to a remote one.
 Needs GGUF files or a running server
 :::
 
+:::{grid-item-card} Read the report
+:link: report
+:link-type: doc
+
+The `Queries` block is text to paste into another model.
++++
+No extra install
+:::
+
 ::::
 
 ```{toctree}
@@ -50,4 +60,5 @@ Needs GGUF files or a running server
 
 pbmc8k_annotation
 load_model
+report
 ```

{annotuneagent-0.1.0 → annotuneagent-0.1.1}/docs/tutorials/load_model.ipynb

@@ -12,7 +12,7 @@
     "| | Use it when |\n",
     "|---|---|\n",
     "| Local | You download the weights and run `llama-server` on this computer |\n",
-    "| Remote | Name and ML each have a running HTTP endpoint |\n",
+    "| Remote | One `llama-server` is already running with both adapters |\n",
     "\n",
     "With `models`, the marker score, the name model, and the ML model each return a result. A judgment reads the three results and writes the title. The title is a dictionary name. The name model copies a cell-type name from the marker dictionary. The ML model is trained on machine-learning measurements of the cluster, differential expression and marker-panel statistics, and copies a candidate name. A cluster can stay unnamed. Without `models`, only the marker score is collected."
    ]
@@ -36,7 +36,9 @@
     "|---|---|\n",
     "| Linux with an NVIDIA GPU | Supported. Steps below. |\n",
     "| macOS (Apple silicon or Intel) | Supported. Use the macOS steps below. The Metal build uses the GPU automatically. |\n",
-    "| Windows | Not supported for the local section. `import annotuneagent` works, but starting and stopping `llama-server` uses POSIX process calls. Run the server on Linux or macOS and use the Remote section, or use WSL. |\n",
+    "| Windows | This package does not start `llama-server`. Use [AnnoTuneLoader](https://github.com/leelieber2025/AnnoTuneLoader). |\n",
+    "\n",
+    "On Windows, download [AnnoTuneLoader.exe](https://github.com/leelieber2025/AnnoTuneLoader/raw/main/AnnoTuneLoader.exe) and double-click it. Choose the base, the name adapter, and the ML adapter. The server listens at `http://127.0.0.1:8080`. Leave the window open while you annotate. On that computer, `local_models(model_root)` reuses the server when those three files match. Another computer uses the Remote section below.\n",
     "\n",
     "**Linux, CUDA 12.8.** Unpack the server and the CUDA runtime next to each other.\n",
     "\n",
@@ -202,7 +204,7 @@
    "source": [
     "## Remote\n",
     "\n",
-    "Use this when name and ML each have a running endpoint. Set the addresses of the two model services. No local GGUF files are required."
+    "Use this when one `llama-server` is already running with both adapters, including one started by [AnnoTuneLoader](https://github.com/leelieber2025/AnnoTuneLoader) on another machine. Point both tasks at that server. Each request calls `/completion` and enables one adapter. No local GGUF files are required. Name prompts must fit in 2048 tokens. ML prompts must fit in 8192. Those limits apply when `maxlen` is omitted."
    ]
   },
   {
@@ -214,11 +216,10 @@
    "source": [
     "import annotuneagent as ata\n",
     "\n",
-    "name_url = \"http://NAME_HOST:8080\"\n",
-    "ml_url = \"http://ML_HOST:8080\"\n",
+    "server = \"http://SERVER:8080\"\n",
     "models = ata.ModelChain(\n",
-    "    name=ata.TaskEndpoint(url=name_url),\n",
-    "    ml=ata.TaskEndpoint(url=ml_url),\n",
+    "    name=ata.TaskEndpoint(url=server),\n",
+    "    ml=ata.TaskEndpoint(url=server),\n",
     ")"
    ]
   },

annotuneagent-0.1.1/docs/tutorials/report.md

@@ -0,0 +1,126 @@
+# Read the report
+
+`run` returns a dictionary and prints a `Queries` block. Each block is one
+cluster that still has a question: it has no title, or the title is written
+and `hint` is set. AnnoTuneAgent does not send this text to the name model,
+the ML model, or any other service. Copy the paragraph under the heading and
+paste it into another model.
+
+## The heading
+
+```text
+cluster 0  monocyte  no marker passed
+```
+
+The three fields are the cluster id, the written title, and the hint. A
+cluster with no title prints `no label` in the middle. The heading stays in
+the notebook. It is not part of the question.
+
+`no marker passed` means the title `monocyte` was written, and none of its
+cited markers passed the expression check. The name stays. The dotplot still
+draws the best-measured cited genes for that type.
+
+```text
+cluster 2  CD14-positive monocyte  conflict: marker neutrophil
+```
+
+The written title is `CD14-positive monocyte`. The marker score named
+`neutrophil`, and that name is not on the same Cell Ontology lineage as the
+title. The hint records the other call. The same line can name the name model
+or the ML model, for example `conflict: ML pericyte`. Several disagreements
+are joined with commas.
+
+## The paragraph to paste
+
+```text
+human, 1728 cells. What cell type is this cluster?
+Enriched versus other cells: HLA-DRA (in 1.00, out 0.62, logfc 2.46); HLA-DRB1 (in 0.99, out 0.65, logfc 2.15); CD74 (in 1.00, out 0.93, logfc 1.84); LGALS2 (in 0.74, out 0.15, logfc 3.40); HLA-DPB1 (in 0.95, out 0.56, logfc 2.08); HLA-DRB5 (in 0.94, out 0.53, logfc 1.96); PID1 (in 0.70, out 0.19, logfc 2.78); CPVL (in 0.84, out 0.36, logfc 2.07); HLA-DPA1 (in 0.96, out 0.58, logfc 1.98); LYZ (in 1.00, out 0.65, logfc 3.01).
+Also detected: CD14 (in 0.87, out 0.42, logfc 1.71); FCN1 (in 0.99, out 0.59, logfc 1.83); CSF1R (in 0.62, out 0.31, logfc 1.16); S100A8 (in 1.00, out 0.57, logfc 2.39); VCAN (in 0.94, out 0.50, logfc 1.59); S100A4 (in 1.00, out 0.95, logfc 0.81); ANXA1 (in 0.81, out 0.69, logfc 0.35).
+```
+
+The first line is the species, the tissue when `run` was given one, the
+number of cells, and the question. Paste from this line through `Also
+detected`.
+
+`Enriched versus other cells` lists up to 10 genes that are higher in this
+cluster than in the other cells, in Wilcoxon order. `in` is the fraction of
+cells in the cluster where the gene is detected. `out` is that fraction in
+the other cells. `logfc` is the Scanpy log fold change.
+
+`Also detected` lists markers of dictionary programs that are on in this
+cluster, detected in at least half of its cells, and not already in the
+enriched list. The list stops at 16 genes.
+
+A reply from the other model is not written back into `adata` or the report.
+Use it when you decide whether to keep or change the title in
+`annotation_code`.
+
+The same paragraph is `query` on `report["clusters"]` and on
+`report["review"]`.
+
+## One row per cluster
+
+`report["clusters"]` has one row per cluster. The columns stay separate. A
+title is not a vote count.
+
+| Column | What it is |
+|---|---|
+| `cluster` | Leiden id |
+| `cells` | Cells in that cluster |
+| `marker` | Name from the marker score, or none |
+| `name` | First name copied by the name model. `failed` when the call failed. The action when the reply named nothing |
+| `ml` | First name copied by the ML model. `escalate` when it abstains. `failed` when the call failed |
+| `label` | Title written onto the cells. Empty when the cluster stays unnamed |
+| `hint` | The question printed after the title. Empty when there is nothing to ask |
+| `query` | The paragraph to paste. Present when the cluster is unnamed or `hint` is set |
+| `reason` | Why the title was written or withheld |
+| `ontology` | Cell Ontology id for the written title. `pending` means the id is not a reviewed row |
+
+`adata.obs["annotune_label"]` is the title. `adata.obs["annotune_hint"]` is
+the hint on every cell of that cluster. `adata.obs["annotune_final_status"]`
+is `named`, `review`, or `escalate`. Clusters that share a title also share
+one id in `adata.obs["annotune_cluster"]`. The Leiden column is kept.
+
+## What to read first
+
+`report["review"]` is the clusters to read first. The order is a reading
+order, not an accuracy score.
+
+| Priority | Which clusters |
+|---|---|
+| 0 | No title |
+| 1 | A conflict, or close marker scores |
+| 2 | A model call failed |
+| 3 | Fewer than 20 cells |
+| 4 | Any other hint |
+
+`report["queue"]` lists only clusters with no title. A titled cluster with a
+hint is in `review`, and on `annotune_hint`, and it is not in `queue`.
+
+## The rest of the report
+
+| Field | What it is |
+|---|---|
+| `records` | The working record for each cluster, including model rounds and the marker ranking |
+| `dropped_hypotheses` | Dictionary types left out because a positive marker is missing from the matrix |
+| `subclusters` | The second split of a cluster that was not settled. Empty when `subcluster=False` |
+| `arbitration` | Clusters that were not given the status `named` |
+| `merged` | Titles that were given one `annotune_cluster` id |
+| `all_labeled` | True when every cell has a title |
+| `written` | True when at least one cell has a title |
+| `label_key` | The `obs` column that holds the title. `None` when nothing was written |
+| `marker_dict` | Genes drawn in the dotplot, in column order |
+| `marker_validation` | Measurement of each cited gene, and why it was kept or left out |
+| `marker_status` | Per written type. `no_marker_passed_expression_check` is the check behind `no marker passed` |
+| `marker_thresholds` | The cutoffs used for that check |
+| `dotplot` | Path of the PDF when `dotplot_path` was set |
+| `doublets` | Doublet summary stored by `prepare` |
+| `tissue` | Tissue page passed to `run` |
+| `tissues` | Tissue pages available for the species |
+| `annotation_code` | Python that redraws the UMAP and the dotplot from the written titles |
+
+`annotation_code` copies the titles into `cluster2celltype`. Clusters with no
+title are left out of that map and listed under `cluster_choices` when the
+record has candidate names. Edit the map or `fine_order`, then run the
+snippet on the annotated object. {doc}`../quickstart` records the expression
+cutoffs used for the dotplot.

{annotuneagent-0.1.0 → annotuneagent-0.1.1}/pyproject.toml

@@ -71,11 +71,12 @@ annotuneagent = "annotuneagent.cli:main"
 
 [project.urls]
 Homepage = "https://github.com/leelieber2025/AnnoTuneAgent"
-Documentation = "https://app.readthedocs.org/projects/annotuneagent/"
+Documentation = "https://annotuneagent.readthedocs.io/"
 Changelog = "https://github.com/leelieber2025/AnnoTuneAgent/blob/main/CHANGELOG.md"
 Source = "https://github.com/leelieber2025/AnnoTuneAgent"
 Issues = "https://github.com/leelieber2025/AnnoTuneAgent/issues"
 PyPI = "https://pypi.org/project/annotuneagent/"
+Zenodo = "https://doi.org/10.5281/zenodo.24092828"
 
 [tool.setuptools.packages.find]
 where = ["src"]

{annotuneagent-0.1.0 → annotuneagent-0.1.1}/src/annotuneagent/__init__.py

@@ -27,9 +27,12 @@ those candidates.
 ``local_models`` takes the directory containing the Q6_K base and the name and
 ml adapters. The tokenizer is installed with the package. The server is
 ``http://127.0.0.1:8080`` unless ``url`` or ``ANNOTUNE_URL`` says otherwise.
-A matching local server is reused. Remote name and ML services use separate
-``TaskEndpoint(url=...)`` values. ``attach`` connects to an existing service
-without loading weights.
+A matching server is reused. When the process list is unavailable, the check
+reads ``/props`` and ``/lora-adapters``. ``TaskEndpoint(url=...)`` calls
+``/completion`` on that server and enables one adapter per request.
+``attach`` connects to an existing service without loading weights.
+Model requests wait 600 seconds unless ``timeout`` or ``ANNOTUNE_TIMEOUT``
+says otherwise.
 """
 
 from __future__ import annotations

{annotuneagent-0.1.0 → annotuneagent-0.1.1}/src/annotuneagent/_version.py

@@ -7,7 +7,7 @@ from __future__ import annotations
 
 __all__ = ["__version__", "version", "version_tuple"]
 
-__version__ = "0.1.0"
+__version__ = "0.1.1"
 version = __version__
 
 _parts: list[int | str] = []

{annotuneagent-0.1.0 → annotuneagent-0.1.1}/src/annotuneagent/decide.py

@@ -628,10 +628,27 @@ def hit_names(payload: dict[str, object]) -> set[str]:
     return {str(item["name"]) for item in payload.get("dictionary_hits") or []}
 
 
-def fetch_name(url: str, payload: dict[str, object], timeout: float = 60) -> Decision:
-    return parse_decision(
-        post_chat(url, SYSTEM_NAME, payload, timeout, task="name"), hit_names(payload)
+def fetch_name(
+    url: str,
+    payload: dict[str, object],
+    timeout: float | None = None,
+    *,
+    maxlen: int = 2048,
+    tokenizer: str | None = None,
+) -> Decision:
+    """Name task on ``/completion``. One adapter is enabled for this request."""
+    from .local_lora import complete_url
+
+    text = complete_url(
+        url,
+        "name",
+        SYSTEM_NAME,
+        payload,
+        maxlen=maxlen,
+        timeout=timeout,
+        tokenizer=tokenizer,
     )
+    return parse_decision(text, hit_names(payload))
 
 
 def candidate_names(payload: dict[str, object]) -> set[str]:
@@ -644,11 +661,27 @@ def candidate_names(payload: dict[str, object]) -> set[str]:
     return names
 
 
-def fetch_ml(url: str, payload: dict[str, object], timeout: float = 120) -> MlResult:
-    return parse_ml(
-        post_chat(url, SYSTEM_ML, payload, timeout, task="ml"),
-        candidate_names(payload),
+def fetch_ml(
+    url: str,
+    payload: dict[str, object],
+    timeout: float | None = None,
+    *,
+    maxlen: int = 8192,
+    tokenizer: str | None = None,
+) -> MlResult:
+    """ML task on ``/completion``. One adapter is enabled for this request."""
+    from .local_lora import complete_url
+
+    text = complete_url(
+        url,
+        "ml",
+        SYSTEM_ML,
+        payload,
+        maxlen=maxlen,
+        timeout=timeout,
+        tokenizer=tokenizer,
     )
+    return parse_ml(text, candidate_names(payload))
 
 
 def _negative_high(row: dict | None) -> bool: