b9478e6a17
batch_size= argument to embed_multi(), refs #273
2023-09-13 16:24:04 -07:00
f54f2c659d
response.__str__ method, closes #268
2023-09-12 10:36:29 -07:00
52cec1304b
Binary embeddings ( #254 )
...
* Binary embeddings support, refs #253
* Write binary content to content_blob, with tests - refs #253
* supports_text and supports_binary embedding validation, refs #253
2023-09-11 18:58:44 -07:00
e49c816e18
Drop llm.EmbeddingModel.embedding_size property, closes #208
2023-09-02 08:45:17 -07:00
ec15707588
next(iter(...)), refs #200
2023-08-31 22:55:58 -07:00
817aff8bae
Embeddings now happen in batches, closes #200
2023-08-31 22:27:06 -07:00
77cf56e54a
Initial CLI support and plugin hook for embeddings, refs #185
...
* Embeddings plugin hook + OpenAI implementation
* llm.get_embedding_model(name) function
* llm embed command, for returning embeddings or saving them to SQLite
* Tests using an EmbedDemo embedding model
* llm embed-models list and emeb-models default commands
* llm embed-db path and llm embed-db collections commands
2023-08-27 22:24:10 -07:00
dff36f0edc
Change of policy: keys.json over-rides environment variables, closes #158
2023-08-20 23:27:21 -07:00
14a91efbad
Fixes for Pydantic 1, including matrix test - refs #169
...
Also refs #147
2023-08-19 20:59:30 -07:00
740a8e8344
Allow for use of Pydantic v1 as well as v2. ( #147 )
...
* Allow for use of Pydantic v1 as well as v2.
Use v1-style configuration. While this is officially
deprecated (https://docs.pydantic.dev/latest/usage/model_config/ ),
it is supported by v2. I would assume this support will remain
for a while, this will help ease the transition to v2
2023-08-19 20:50:59 -07:00
cb41409e2b
conversation_name should not have newlines, closes #110
2023-07-15 21:28:35 -07:00
19ba1a3099
Improvements to Response.fake() plus __repr__ for some classes
2023-07-11 15:14:52 -07:00
abfc9c69dc
Response.fake() method
...
I'm using this in https://github.com/simonw/llm-gpt4all/issues/4
2023-07-11 11:08:15 -07:00
18a89d8bb8
Removed obsolete OptionsError class
2023-07-11 10:53:04 -07:00
255f768707
Rename ._response_json to .response_json
2023-07-11 08:31:38 -07:00
2d3ebe7fe1
llm logs now uses new DB schema, refs #91
2023-07-11 07:21:22 -07:00
88d839fa85
Create conversations and responses tables and write to them, refs #91
2023-07-11 07:21:22 -07:00
21e6d54365
Remove unused imports
2023-07-11 00:09:35 -07:00
23eeb0f322
Remove experimental Model.chain() method, refs #90
2023-07-10 22:47:34 -07:00
b38b8314b9
Initial working implementation of conversations, refs #85
2023-07-10 22:37:45 -07:00
199f7e0767
Renamed iter_prompt() to execute() and updated tutorial
2023-07-10 07:59:12 -07:00
ae87f978bd
Moved iter_prompt from Response to Model, moved a lot of other stuff
...
- Moved a whole bunch of things from llm/cli.py into llm/__init__.py
- Switched plugin listings to use importlib.metadata to avoid deprecation warning
- iter_prompt() is now a method on Model, not on Response
2023-07-10 07:45:11 -07:00
902cf6010c
Fixed type hint on Prompt
2023-07-07 20:38:53 -07:00
04568115b8
Options base class is now llm.Options not Model.Options
2023-07-05 20:47:36 -07:00
4611bff412
iter_prompt() now takes prompt
2023-07-05 20:46:17 -07:00
b88906d459
Default __str__ method for models
2023-07-05 18:26:16 -07:00
6ef6b343a9
Improved how keys work, execute() now has default implementation
2023-07-05 18:25:57 -07:00
78c93b9e23
Model.execute() model now defaults to using self.Response
2023-07-05 16:57:10 -07:00
bc8dad1e33
stream defaults to True on prompt() method
2023-07-04 08:14:00 -07:00
49e4d688c6
Removed .stream() method in favor of .prompt(stream=False)
2023-07-04 07:50:31 -07:00
3136948408
Moved things into inner classes, log_message is now defined on base Response
2023-07-03 21:25:19 -07:00
345ad0d2dc
Implemented new logs database schema
2023-07-03 07:27:47 -07:00
b1c51df3f1
New LogMessage design, plus Response.json() method
2023-07-03 06:46:51 -07:00
61dd8afc60
Drop the debug field from the logs, combine chunks from stream
2023-07-03 06:39:54 -07:00
84b99f8baf
-o/--option, implemented for OpenAI models - closes #63
2023-07-02 17:42:22 -07:00
52add96ec1
Rough initial version of new logging, to log2 table
2023-07-02 16:35:36 -07:00
3ed86e403d
Added mypy, plus some fixes to make it happy - refs #77
2023-07-02 12:36:22 -07:00
7054c673e1
Better type hint for iter_prompt method
...
The Generator thing is only useful if the method uses the send mechanism
Declaring it an Iterator gives people more flexibility for how they implement it.
2023-07-02 11:18:31 -07:00
f3f38c0c02
Initial experimental response.reply() method
2023-07-01 13:29:54 -07:00
664fca165b
.stream is now in base Response class
2023-07-01 13:29:27 -07:00
9ac120fae4
Upgrade to pydantic 2 using bump-pydantic, refs #74
2023-07-01 11:53:54 -07:00
9afc758cd7
Pass model to the Response
2023-07-01 11:29:41 -07:00
ffe4b6706d
Model.stream() and .get_key() methods
2023-07-01 09:03:07 -07:00
5e056fad8a
Implemented PaLM 2, to test out new plugin hook - refs #20
2023-06-26 08:25:01 -07:00
5103f77c40
New mechanism for handling API keys for models
2023-06-26 08:05:48 -07:00
274126d9f8
WIP register_models() plugin hook, refs #53
2023-06-26 07:43:46 -07:00
Simon Willison