plyllm 1.0.0__tar.gz

plyllm-1.0.0/.claude/skills/update-models/SKILL.md

@@ -0,0 +1,52 @@
+---
+name: update-models
+description: Update supported model definitions using current official provider documentation.
+---
+
+Read `references/sources.md` before step 2. For every provider it gives the exact official
+documentation URLs to fetch, the repository's conventions for that provider (which pricing tier,
+the cache-read multiplier, how thinking and effort are encoded), and the places where a provider's
+own pages contradict each other. It is a shortcut past the search, not a substitute for the docs:
+always re-fetch the pages and verify against them. Update `references/sources.md` in the same change
+whenever a URL moves, a provider is added, or a convention changes.
+
+For every model provider supported by this repository:
+
+1. Inspect the repository to find the provider's model definitions and related metadata.
+2. Check the provider's official developer documentation for:
+   - current API model IDs
+   - pricing in USD
+   - reasoning/thinking support and available settings
+   - deprecated, renamed, or newly available models
+
+3. Update the repository with the latest verified information.
+   - Preserve existing data structures, naming conventions, and pricing units.
+   - Add new supported models and update outdated entries.
+   - Remove or deprecate models only when appropriate for the repository.
+   - Do not guess undocumented pricing or capabilities.
+
+4. Update related aliases, tests, fixtures, or documentation when required.
+5. Run the relevant tests, formatter, and type checks. Then run `make test-models`: it is free (each
+   provider's model-list endpoint, keys from .env) and compares every active registry model with the
+   ids the provider serves right now. Read a failure before changing anything:
+   - `no longer serves [...]`: first check whether the id is an alias the listing omits (Anthropic
+     lists only the dated snapshot of some models, e.g. `claude-haiku-4-5-20251001`). If so, switch
+     `model_string` to the listed id the docs say it resolves to. Mark a model `retired=True` only
+     when the docs or the listing show it is really gone; never delete it, recorded results need it.
+   - An error raised while listing (401, 403, "used all available credits", a timeout) is an
+     account or network problem, not a registry one. Report it and leave the registry alone.
+   - A warning that a retired model is served again: report it; do not un-retire on your own.
+
+   `make test-models` only checks models already in the registry; it cannot see one that was left
+   out. `references/sources.md` lists models that were removed or skipped because the account could
+   not use them (e.g. grok-4.7 / grok-4.6), each with a free command to check whether they are served
+   now. Run those checks too, and re-add a model as soon as it is served.
+
+6. Summarize:
+   - providers checked
+   - models added, updated, or removed
+   - pricing or reasoning changes
+   - validation performed, including the `make test-models` result per provider
+   - official documentation used
+
+Use official provider documentation as the source of truth. Research every provider supported by the repository, not only providers such as OpenAI, Gemini, or Anthropic.

plyllm-1.0.0/.claude/skills/update-models/references/sources.md

@@ -0,0 +1,198 @@
+# Provider documentation sources
+
+Where the model data in `src/plyllm/providers/*/models.py` comes from. Verified 2026-10-09.
+
+Fetch the pricing page and the per-model pages in parallel; the pricing pages carry every model's
+rates in one table, and the per-model pages are the only place that lists a model's exact set of
+reasoning/effort levels. A general "reasoning" guide and a per-model page sometimes disagree about
+which levels a model accepts — prefer the intersection, because an unsupported level is a runtime
+400 while a missing one only blocks a config.
+
+Batch prices (checked 2026-10-10) are a `batch_ratio` on each model, applied to input, cached input and
+output: OpenAI 0.5 (every Batch row is half the Standard row; gpt-5.4 cached shows $0.13, rounded), Claude
+0.5 (stacks with the cache multipliers), Mistral 0.5 (whether it covers cached input is not stated),
+Gemini 0.5 (see its section for cache reads), Grok per model (0.8 on grok-4.3 and grok-4.20; no discount
+listed for the others, which get `None`), Metacentrum `None` (free). Batch: OpenAI
+<https://developers.openai.com/api/docs/guides/batch>, Anthropic pricing page, Gemini
+<https://ai.google.dev/gemini-api/docs/batch-mode>, xAI <https://docs.x.ai/developers/pricing>, Mistral
+<https://mistral.ai/pricing>.
+
+## Anthropic (`claude`)
+
+- Pricing (input / output / cache read + write, batch, all models): <https://platform.claude.com/docs/en/about-claude/pricing>
+- Model lineup, IDs, context, max output, thinking mode, default effort: <https://platform.claude.com/docs/en/about-claude/models/overview>
+- Effort levels per model, and per-model recommendations: <https://platform.claude.com/docs/en/build-with-claude/effort>
+- One model's full spec: `https://platform.claude.com/docs/en/models/<slug>/overview` (e.g. `opus-5-5`, `fable-5-1`)
+- Deprecations / retirements: <https://platform.claude.com/docs/en/about-claude/model-deprecations>
+
+`docs.claude.com/en/docs/...` 302-redirects to `platform.claude.com/docs/en/...`; fetch the
+`platform.claude.com` URL directly to save a round trip.
+
+Repository notes: `cache_write_cost` is the 5-minute cache-write rate, 1.25x input on every model
+(`_CACHE_WRITE_RATIO`); the client caches the system prompt with the default 5-minute lifetime, so the
+1-hour rate (2x) is not stored. `cached_input_cost` is the cache-read rate, normally 0.1x input but 0.025x on
+Fable 5.1 and 0.05x on Opus 5.5 and Sonnet 5.5. `cache_write_cost` is the 5-minute cache-write rate,
+1.25x input on every model (`_CACHE_WRITE_RATIO`); the client caches the system prompt with the default
+5-minute lifetime, so the 1-hour rate (2x) is never charged. The pricing page says the batch discount
+stacks with both cache multipliers. The models endpoint lists some older models only by dated snapshot (`claude-haiku-4-5-20251001`,
+`claude-opus-4-5-20251101`, `claude-sonnet-4-5-20250929`), not by alias, so use the snapshot there. Haiku 5.5 has two price tiers (prompts up to / over
+100k tokens); the repository stores the lower one. "Adaptive (always on)" in the docs maps to `thinking_only=True`.
+Models whose docs say thinking cannot be disabled at any effort are `thinking_only`; Opus 5 can only
+disable it at `high` or below, which `_THINKING_ONLY_EFFORTS` already encodes. Sonnet 5.5 rejects
+`"disabled"` and only offers `thinking: {"type": "between_tools"}` (effort `high` or below), which the
+harness does not model, so it is `thinking_only`. The per-model thinking table ("Rejected with 400")
+is at <https://platform.claude.com/docs/en/build-with-claude/thinking-troubleshooting>. Its footnote
+names only Opus 5 and Haiku 5.5 as rejecting `"disabled"` at `xhigh`/`max`; the repository applies
+that rule to every Claude model (`_THINKING_ONLY_EFFORTS`), which the docs neither confirm nor rule
+out for Opus 4.8, Sonnet 5 and Sonnet 4.6.
+
+## OpenAI (`openai`)
+
+- Pricing, all models, input / cached input / output: <https://developers.openai.com/api/docs/pricing>
+- Current model list: <https://developers.openai.com/api/docs/models>
+- Per-model effort values, default and pricing: `https://developers.openai.com/api/docs/models/<model-id>`
+- Reasoning guide (general, does not enumerate per model): <https://developers.openai.com/api/docs/guides/reasoning>
+
+`platform.openai.com/docs/...` 301-redirects to `developers.openai.com/api/docs/...`.
+
+Repository notes: the per-model pages list `none` as an effort value; the harness expresses that as
+`thinking_enabled=False` (the provider sends effort `none`), so `none` is never in a `supported_reasoning`
+set. A model whose page does not list `none` is `thinking_only` (on 2026-10-09: gpt-6-astra, gpt-6.1-sol,
+the Pro models, and gpt-5-mini / gpt-5-nano, whose pages list no efforts at all). A model whose default
+effort is `none` (gpt-5.4, -mini, -nano) is `needs_effort_to_think`, or thinking on would not reason.
+Temperature is accepted only at effort `none` (GPT-6 migration guide, "When reasoning effort is not none,
+remove temperature, top_p"), which `OpenAILLMModel` derives from `thinking_only`. Effort sets differ
+within a generation — GPT-6.x and GPT-5.6 accept `max`, GPT-5.5 and GPT-5.4 stop at `xhigh`, and the
+Pro variants start at `medium`. Cached input is not always 0.1x: gpt-6.1-sol is 0.05x. gpt-6.1-sol rejects `none` (thinking cannot be
+turned off), and the gpt-6-astra page does not list `none`. gpt-5.6-sol is on promotional pricing "at
+least through November 21, 2026". The gpt-5-mini / gpt-5-nano pages no longer list effort values.
+Shutdown dates: <https://developers.openai.com/api/docs/deprecations>.
+
+## Google Gemini (`gemini`)
+
+- Pricing, paid tier, per model: <https://ai.google.dev/gemini-api/docs/pricing>
+
+`cached_input_cost` is the "Context caching price" row (hourly storage is not modelled). The batch table
+lists its own cache-read price; where it is not half of the standard one (2.5 models, 3.1 Pro, 3 Flash:
+same as standard; 3.5 Flash-Lite: 0.02), record it as `batch_cached_input_cost`. The batch-mode page says
+batch cache hits pay "the standard context caching rates", which contradicts the table for 3.8 / 3.6 Flash
+and 3.1 Flash-Lite; the table is used. 3.7 Flash and 3.5 Flash had no pricing block on 2026-10-10, so
+their cache price is 0 until the page lists one.
+
+- Model list, stable vs preview: <https://ai.google.dev/gemini-api/docs/models>
+- Deprecations and shutdown dates: <https://ai.google.dev/gemini-api/docs/deprecations>
+- Embedding task prefixes (`task:` / `title:`), verbatim: <https://ai.google.dev/gemini-api/docs/embeddings>
+
+Repository notes: Flash pricing is promotional and dated — 3.8 / 3.7 / 3.6 Flash are $0.75 / $3.75
+through 2026-12-31 and $1.50 / $7.50 after. Record the rate in force, with a comment giving the
+change date. Pro models have a >200k-token tier; the repository stores the standard tier only.
+The models page lists preview models that the deprecations page has already shut down — always
+cross-check both before keeping a `-preview` id (on 2026-10-09 it still listed
+`gemini-embedding-2-preview`, which shut down on 2026-08-10). The thinking guide
+(<https://ai.google.dev/gemini-api/docs/thinking>) is the per-model table of thinking levels and
+defaults; it no longer documents `thinking_budget`, which the 2.5 models still use here. 3.8 and 3.7
+Flash reject `minimal`. 3.1 Flash-Lite is not in that table.
+
+## xAI Grok (`grok`)
+
+- Model list and pricing table: <https://docs.x.ai/docs/models>
+- Per-model page, with the effort list and default: `https://docs.x.ai/docs/models/<model-id>`
+- Reasoning guide: <https://docs.x.ai/docs/guides/reasoning>
+
+Repository notes: prices are the standard sub-200k-token tier; requests reaching 200k are billed at
+roughly double for the whole request. The reasoning guide and the per-model pages disagree about
+`xhigh` on grok-4.5 (the guide says it is silently treated as `high`) and about whether grok-4.3
+reasons at all — both are kept at low/medium/high. The guide documents `reasoning_effort` only for
+grok-4.7 / 4.6 / 4.5, and the grok-4.20 page has no effort section, so grok-4.20 takes none.
+Reasoning cannot be disabled on grok-4.7 / 4.6 / 4.5, so they are `thinking_only`, as is grok-4.20
+(no effort section; it always reasons); grok-4.3 accepts `none` (default `low`), which the provider sends
+when thinking is off. No page documents `temperature` for the reasoning models, so every Grok model has
+`TemperatureSupport.NEVER`; nor is `reasoning.summary` documented (it is still sent with an effort). Retirements: <https://docs.x.ai/developers/migration/may-15-retirement>.
+
+**Waiting to be re-added: grok-4.7 and grok-4.6.** Removed on 2026-10-09: the docs list them, but the
+API returns 404 for this account ("does not exist or your team does not have access to it"). On every
+update, check whether they are served now (free, needs `GROK_API_KEY`):
+
+```
+uv run python -c "import asyncio; from plyllm import LLM, LLMConfig, Provider; print(sorted(asyncio.run(LLM(LLMConfig.from_env()).served_models(Provider.GROK))))"
+```
+
+If the list contains them, add them back with the values from their model pages (on 2026-10-09:
+$2.00 / $6.00, cached $0.50; effort low/medium/high/xhigh, default high; reasoning cannot be
+disabled), run `make test-models`, and delete this paragraph.
+
+## Mistral (`mistral`)
+
+- API pricing per model: <https://docs.mistral.ai/inference/pricing> (`mistral.ai/pricing/api` 301-redirects here)
+- One model's page, with its API id: `https://docs.mistral.ai/models/<slug>` (e.g. `mistral-large-4-0`,
+  `mistral-medium-3-5-26-04`, `mistral-small-4-0-26-03`; the undated slugs of the last two 404)
+- Model list, dated ids, deprecations: <https://docs.mistral.ai/getting-started/models/models_overview/>
+- Reasoning capability: <https://docs.mistral.ai/capabilities/reasoning/>
+
+Repository notes: docs slugs are not API ids — the model page shows the id next to the parameter
+count (`mistral-large-2512 +1` for Large 3, `mistral-large-4 +1` for Large 4). The reasoning page
+lists which models take `reasoning_effort` (Small, Medium 3.5, Large 4, hosted GLM 5.3), but only
+documents `high` and `none`, so `_MISTRAL_REASONING` is `{high}` (`none` is what the provider sends
+when thinking is off). Do not take the set from the SDK enum (`none, minimal, low, medium, high,
+xhigh`): on 2026-10-09 `mistral-small-2603` rejected `low` with "Must be one of (none, high)".
+Widen the set for a model only after `make test-live` (or a single paid call, approved first) shows
+the API accepts the level. Cached input is documented only as "up to -90% on input tokens", so `cached_input_cost` is
+0.1x input. Large 4 is on an undated sale; the repository records the sale price
+with the list price in a comment. Large 3 and Ministral 3 are deliberately not in the repository
+because their reasoning support is undocumented. Hosted GLM 5.3 (`zai-glm-5-3`) documents only
+`low`/`high`/`max` and rejects `none`, which the provider sends when thinking is off; it is left out
+until the provider can express that. The Medium 3.5 page lists `mistral-medium-3-5`,
+`mistral-medium-3` and `mistral-medium-latest`, not the `mistral-medium-2604` the repository sends;
+`make test-models` shows whether the endpoint still serves it. WebFetch has swapped price labels on
+the Large 4 page; read the pricing table from raw HTML (`curl`) when in doubt.
+
+## Metacentrum / e-INFRA (`metacentrum`)
+
+- Served models, API ids, aliases, obsolete models: <https://docs.cerit.io/en/docs/ai-as-a-service/chat-ai>
+- API base URL: `https://llm.ai.e-infra.cz/v1/` (set locally through `METACENTRUM_BASE_URL` / `METACENTRUM_API_KEY`)
+
+Repository notes: self-hosted and free, so every model carries `input_cost=0` / `output_cost=0`.
+The service replaces models without notice and the page has a separate "Obsolete Models" table —
+check it, not just the main table. The page can lag the live endpoint (on 2026-10-07 it still read
+"Effective June 30, 2026" and listed `deepseek-v4-flash`, while `/v1/models` served
+`deepseek-v4.1-flash` from 2026-09-28). Prefer `GET https://llm.ai.e-infra.cz/v1/models` (free,
+needs `METACENTRUM_API_KEY`) and ask before calling it. Models it drops are **kept** in `models.py`
+and marked `retired=True`, so existing experiment configs and recorded results still resolve and
+cost (`resolve_model` raises on an unknown name) while new calls are refused. `make test-models`
+(free, needs the keys in .env) compares every provider's active registry models with its served list
+and fails on a mismatch. Reasoning-effort support is not documented per model, so leave
+`supported_reasoning=None` unless a model is known to accept it; the only documented toggle is
+DeepSeek's `chat_template_kwargs: {"thinking": true}` (reasoning is off by default), on
+<https://docs.cerit.io/en/docs/ai-as-a-service/ai-api>. That page is out of date: on 2026-10-09
+deepseek-v4.1-flash reasoned with and without it, and with `thinking: false`. The thinking flags come
+from a free probe instead (one short prompt per model, thinking on and off, reasoning tokens counted; put
+a unique id in each prompt, because the proxy caches identical prompts and ignores `extra_body` when it
+does). Results on 2026-10-09: the Qwen templates read `enable_thinking` (not `thinking`), kimi-k3 reads
+`thinking`, gpt-oss-120b and deepseek-v4.1-flash always reason (`thinking_only`), gemma-4 reasons only with
+an effort (`needs_effort_to_think`), mistral-medium-3.5 gave no reasoning with thinking on
+(`thinking=False`), and glm-5.3 answered every call with 400 "`tools` must not be an empty array". On
+2026-10-10 that happened even for a minimal `/v1/responses` request, while `/v1/chat/completions` answered
+normally: the proxy's Responses-to-chat conversion breaks for this model, not the harness's request. The stable aliases (`qwen3.5`, `kimi`,
+`glm`, ...) resolve to whatever is current, so they keep configs working but not outputs: `qwen3.5`
+points at `qwen3.5-int4` (397B, AWQ int4).
+
+## Hugging Face (`huggingface`)
+
+Local encoder checkpoints run through `transformers`; they are free, have no API pricing and no
+reasoning settings. Only verify the repo id still exists, e.g.
+<https://huggingface.co/princeton-nlp/sup-simcse-bert-base-uncased>.
+
+## After editing
+
+```
+make format
+make test
+make typecheck
+make test-models   # free: are all active models still served? Needs the keys in .env
+```
+
+Then check the prices and effort sets round-trip:
+
+```
+uv run python -c "from plyllm.providers.claude.models import claude_models; print([(m.model_name, m.input_cost, m.output_cost) for m in claude_models()])"
+```

plyllm-1.0.0/.env.example

@@ -0,0 +1,28 @@
+# openai configuration
+OPENAI_ORGANIZATION=
+OPENAI_PROJECT=
+OPENAI_API_KEY=
+
+# gemini configuration
+GEMINI_PROJECT=
+GEMINI_API_KEY=
+
+# grok configuration
+GROK_API_KEY=
+GROK_BASE_URL=
+
+# claude configuration
+CLAUDE_API_KEY=
+
+# mistral configuration
+MISTRAL_API_KEY=
+
+# metacentrum configuration
+METACENTRUM_BASE_URL=
+METACENTRUM_API_KEY=
+
+# huggingface configuration (only needed for gated/private models)
+HF_TOKEN=
+
+# requests that share this key and a prompt prefix reuse the provider cache (OpenAI, Grok, Mistral); default PlyLLM
+PROMPT_CACHE_KEY=

plyllm-1.0.0/.github/workflows/ci.yml

@@ -0,0 +1,85 @@
+name: CI
+
+on:
+  push:
+    branches: [master]
+  pull_request:
+
+concurrency:
+  group: ci-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  lint:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v5
+        with:
+          enable-cache: true
+
+      - name: Sync dependencies
+        run: uv sync --group dev --python 3.13
+
+      - name: Ruff lint
+        run: uv run ruff check .
+
+      - name: Ruff format check
+        run: uv run ruff format --check .
+
+      - name: Prettier format check
+        run: npx --yes prettier@3.9.6 --check .
+
+  typecheck:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v5
+        with:
+          enable-cache: true
+
+      - name: Sync dependencies
+        run: uv sync --group dev --python 3.13
+
+      - name: Pyright
+        run: make typecheck
+
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: ["3.12", "3.13"]
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v5
+        with:
+          enable-cache: true
+
+      - name: Sync dependencies
+        run: uv sync --group dev --python ${{ matrix.python-version }}
+
+      - name: Run tests
+        run: uv run --group dev --python ${{ matrix.python-version }} pytest -q
+
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v5
+        with:
+          enable-cache: true
+
+      - name: Build wheel and sdist
+        run: uv build
+
+      - name: Check package metadata
+        run: uvx twine check dist/*

plyllm-1.0.0/.github/workflows/publish.yml

@@ -0,0 +1,57 @@
+name: Publish
+
+# Publishes to PyPI when a GitHub Release is published (see scripts/release.py).
+# Uses PyPI Trusted Publishing (OIDC) — no API token is stored.
+# One-time setup on PyPI: add a trusted publisher for project "plyllm" with
+#   owner=radajakub, repository=plyllm, workflow=publish.yml, environment=pypi.
+on:
+  release:
+    types: [published]
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v5
+
+      - name: Verify tag matches the package version
+        run: |
+          version="$(uv version --short)"
+          tag="${GITHUB_REF_NAME#v}"
+          if [ "$version" != "$tag" ]; then
+            echo "tag $GITHUB_REF_NAME does not match pyproject version $version" >&2
+            exit 1
+          fi
+
+      - name: Build wheel and sdist
+        run: uv build
+
+      - name: Check package metadata
+        run: uvx twine check dist/*
+
+      - name: Upload build artifacts
+        uses: actions/upload-artifact@v4
+        with:
+          name: dist
+          path: dist/
+
+  publish:
+    needs: build
+    runs-on: ubuntu-latest
+    environment:
+      name: pypi
+      url: https://pypi.org/p/plyllm
+    permissions:
+      id-token: write # required for trusted publishing
+    steps:
+      - name: Download build artifacts
+        uses: actions/download-artifact@v4
+        with:
+          name: dist
+          path: dist/
+
+      - name: Publish to PyPI
+        uses: pypa/gh-action-pypi-publish@release/v1

plyllm-1.0.0/.gitignore

@@ -0,0 +1,220 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[codz]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#   Usually these files are written by a python script from a template
+#   before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py.cover
+*.lcov
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+# Pipfile.lock
+
+# uv
+#   Similar to Pipfile.lock, it is generally recommended to include uv.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+# uv.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+# poetry.lock
+# poetry.toml
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#   pdm recommends including project-wide configuration in pdm.toml, but excluding .pdm-python.
+#   https://pdm-project.org/en/latest/usage/project/#working-with-version-control
+# pdm.lock
+# pdm.toml
+.pdm-python
+.pdm-build/
+
+# pixi
+#   Similar to Pipfile.lock, it is generally recommended to include pixi.lock in version control.
+# pixi.lock
+#   Pixi creates a virtual environment in the .pixi directory, just like venv module creates one
+#   in the .venv directory. It is recommended not to include this directory in version control.
+.pixi/*
+!.pixi/config.toml
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule*
+celerybeat.pid
+
+# Redis
+*.rdb
+*.aof
+*.pid
+
+# RabbitMQ
+mnesia/
+rabbitmq/
+rabbitmq-data/
+
+# ActiveMQ
+activemq-data/
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.envrc
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs/Zensical documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#   JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#   be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#   and can be added to the global gitignore or merged into this file.  For a more nuclear
+#   option (not recommended) you can uncomment the following to ignore the entire idea folder.
+# .idea/
+
+# Abstra
+#   Abstra is an AI-powered process automation framework.
+#   Ignore directories containing user credentials, local state, and settings.
+#   Learn more at https://abstra.io/docs
+.abstra/
+
+# Visual Studio Code
+#   Visual Studio Code specific template is maintained in a separate VisualStudioCode.gitignore that
+#   can be found at https://github.com/github/gitignore/blob/main/Global/VisualStudioCode.gitignore
+#   and can be added to the global gitignore or merged into this file. However, if you prefer, you
+#   could uncomment the following to ignore the entire vscode folder
+# .vscode/
+# Temporary file for partial code execution
+tempCodeRunnerFile.py
+
+# Ruff stuff:
+.ruff_cache/
+
+# PyPI configuration file
+.pypirc
+
+# Marimo
+marimo/_static/
+marimo/_lsp/
+__marimo__/
+
+# Streamlit
+.streamlit/secrets.toml

plyllm-1.0.0/.prettierignore

@@ -0,0 +1,6 @@
+.venv
+dist
+build
+node_modules
+*.lock
+uv.lock

plyllm-1.0.0/.prettierrc

@@ -0,0 +1,4 @@
+{
+  "tabWidth": 2,
+  "printWidth": 180
+}

plyllm-1.0.0/.python-version

@@ -0,0 +1 @@
+3.13