First revision: DeepSeek, numeric questions, background assistant, UI fixes
All checks were successful
Verify and publish V2 container / build-and-push (push) Successful in 1m2s
All checks were successful
Verify and publish V2 container / build-and-push (push) Successful in 1m2s
Providers and pricing - Add a DeepSeek provider: thinking mode with low/high/max effort, reasoning_content replayed between tool steps, cache-hit usage, /models discovery and catalog prices. - Fix "does not support tool calling": a rejected forced tool_choice (DeepSeek thinking mode) now asks for the call in the prompt; only a genuine no-tools 400 is "unsupported". - Seven reasoning-effort levels (Off to Max) mapped per adapter. - Cached-input price per model, cache-write and DeepSeek off-peak aware cost estimates. - Built-in bge-small embedding model (transformers.js) as the semantic search default. Quizzes - Numeric (type-the-answer) questions: units ignored, optional ± tolerance; editor, runner, importer, templates and generation prompts. - Quiz start screen instead of showing answers; navigator ring no longer clipped and key legend removed; clock only for timed exams; 7-point letter grades. Assistant and generation - Assistant turns run on the server independently of the browser, with re-attach and an explicit stop; drafts tray shows generations in progress. - Per-chat model switcher, @-tagging of class materials, full-width assistant page and a smaller assistant button. - Editable generation instructions and built-in presets, separate from the chatbot import instructions. Other - Dialog footer padding, keyboard hints removed, copy buttons work over plain HTTP, class and folder shares include documents. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
parent
ccdad9728b
commit
f0aabe4d5d
63 changed files with 3892 additions and 488 deletions
21
docs/AI.md
21
docs/AI.md
|
|
@ -5,8 +5,20 @@ discover models where supported (or add a custom model), enable models and selec
|
|||
for generation, chat, explanations and vision. Output limits, capabilities, pricing and
|
||||
reasoning settings can be edited per model. Catalog prices are estimates, not billing records.
|
||||
|
||||
Adapters use OpenAI Responses, Anthropic Messages, Gemini generateContent, OpenAI-compatible
|
||||
Chat Completions and Ollama's native chat API. Configure the endpoint as shown in Settings.
|
||||
Adapters use OpenAI Responses, Anthropic Messages, Gemini generateContent, DeepSeek and
|
||||
OpenAI-compatible Chat Completions, and Ollama's native chat API.
|
||||
|
||||
Reasoning effort has seven levels (Off, Minimal, Low, Medium, High, Extra high, Max); each
|
||||
adapter maps them onto the nearest value its API accepts. DeepSeek accepts low / high / max
|
||||
and turns thinking off at "Off". DeepSeek rejects forced tool calls in thinking mode, so
|
||||
structured generation keeps thinking on and asks for the tool call in the prompt; answer keys
|
||||
were noticeably less reliable with thinking off. Its `reasoning_content` is replayed between
|
||||
tool steps, as the API requires.
|
||||
|
||||
Cost estimates split input into fresh, cache-hit and cache-write tokens. A model's "Cached
|
||||
input price" is used when set; otherwise the provider's published discount applies (10% of
|
||||
input for GPT-5+, Claude and Gemini; DeepSeek's catalog cache-hit price). DeepSeek's
|
||||
off-peak half price is applied by time of call. Configure the endpoint as shown in Settings.
|
||||
Local models are reached **from the server/container**: `localhost` inside Docker is the
|
||||
container itself, not your workstation or Unraid host. Use a reachable LAN address instead.
|
||||
|
||||
|
|
@ -31,7 +43,10 @@ warning on the document. At most 600 pages are read per file.
|
|||
|
||||
Settings → AI → Semantic search chooses an embeddings provider and model (suggested:
|
||||
OpenAI `text-embedding-3-small`, Gemini `gemini-embedding-001`, Ollama `nomic-embed-text`;
|
||||
Anthropic has no embeddings API). Every document passage is embedded in the background and
|
||||
Anthropic and DeepSeek have no embeddings API). Until something else is chosen, the built-in
|
||||
model is used: BAAI bge-small-en-v1.5 (8-bit ONNX, 384 dimensions) running on the server's CPU
|
||||
through transformers.js. It is downloaded once (about 35 MB) from huggingface.co into
|
||||
`DATA_DIR/models` on first use and costs nothing. `LOCAL_EMBEDDINGS=off` disables the default. Every document passage is embedded in the background and
|
||||
stored as a vector in SQLite (`chunk_vectors`). Search, the assistant's document search and
|
||||
focused generation then merge keyword (FTS5) and meaning-based rankings with reciprocal-rank
|
||||
fusion, so exact terms still match while paraphrased questions find the right passage. If the
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue