Commit Graph
730 Commits
Author SHA1 Message Date
Keegan George 7dee59c97f DEV: Remove no longer needed seeded model check 2025-07-21 08:08:20 -07:00
Keegan George eb93e1736b Final spec fix 🤞 2025-07-17 14:46:15 -07:00
Keegan George cc34882074 FIX: rest of specs 2025-07-17 14:34:29 -07:00
Keegan George 4b03cdcc96 FIX: LLM controller spec 2025-07-17 13:31:26 -07:00
Keegan George 08d2f3ddf9 FIX: spam 2025-07-17 11:36:19 -07:00
Keegan George 7af3ce820d DEV: Remove references to translation model 2025-07-17 09:27:08 -07:00
Keegan George 5841543bd9 FIX: validator 2025-07-16 14:51:32 -07:00
Keegan George 5ae80645bb Merge branch 'main' into default-llm-model 2025-07-16 12:07:10 -07:00
Keegan George b675c4c39b DEV: Remove custom prefix in specs 2025-07-16 11:39:55 -07:00
Keegan George 0fadf1da1a DEV: update enumerator 2025-07-16 10:56:18 -07:00
Roman Rizzi 06743d1939 FIX: Fix embeddings to use the old OpenAI tokenizer (#1506) 2025-07-15 14:44:11 -03:00
Keegan George 88811f97bf fixes 2025-07-15 10:23:58 -07:00
Keegan George 2377b286dd DEV: rely on default llm model in spec 2025-07-15 09:56:24 -07:00
Roman Rizzi 6a4dbd8126 FEATURE: Use Personas in Automation's llm_report script (#1499) 2025-07-14 12:47:21 -03:00
Keegan George 8da4318021 DEV: assign_fake_provider 2025-07-11 12:32:36 -07:00
Keegan George b6a88ff03e spec 2025-07-11 12:02:50 -07:00
Roman Rizzi 89bcf9b1f0 FIX: Process succesfully generated embeddings even if some failed (#1500) 2025-07-10 17:51:01 -03:00
Martin Brennan 6b5ea38644 DEV: Improve AI bot conversation submit upload handling (#1497)
Try fix a flaky spec in /ai_bot/homepage_spec.rb by using ember
data rather than inspecting the DOM directly to see if there
are any in-progress uploads.

Also add missing translation for in progress uploads warning.
2025-07-10 17:06:39 +10:00
Natalie Tay d54cd1f602 DEV: Normalize locales that are similar (e.g. en and en_GB) so they do not get translated (#1495)
This commit
- normalizes locales like en_GB and variants to en. With this, the feature will not translate en_GB posts to en (or similarly pt_BR to pt_PT)
- consolidates whether the feature is enabled in `DiscourseAi::Translation.enabled?`
- similarly for backfill in  `DiscourseAi::Translation.backfill_enabled?`
  - turns off backfill if `ai_translation_backfill_max_age_days` is 0 to keep true to what it says. Set it to a high number to backfill everything
2025-07-09 22:21:51 +08:00
Natalie Tay 56f025cf44 FIX: Localize description excerpts as they have limits (#1490) 2025-07-08 10:36:41 +08:00
Rafael dos Santos Silva 6247906c13 FEATURE: Seamless embedding model upgrades (#1486) 2025-07-04 16:44:03 -03:00
Sam ab5edae121 FIX: make AI helper more robust (#1484)
* FIX: make AI helper more robust

- If JSON is broken for structured output then lean on a more forgiving parser
- Gemini 2.5 flash does not support temp, support opting out
- Evals for assistant were broken, fix interface
- Add some missing LLMs
- Translator was not mapped correctly to the feature - fix that
- Don't mix XML in prompt for translator

* lint

* correct logic

* simplify code

* implement best effort json parsing direct in the structured output object
2025-07-04 14:47:11 +10:00
Natalie Tay 2b9a4f9232 FIX: Ignore captions and quotes when detecting locale and update prompts (#1483)
A more deterministic way of making sure the LLM detects the correct language (instead of relying on prompt to LLM to ignore it) is to take the cooked and remove unwanted elements.

In this commit 
- we remove quotes, image captions, etc. and only take the remaining text, falling back to the unadulterated cooked
- and update prompts related to detection and translation
- /152465/12
2025-07-03 22:57:48 +08:00
Rafael dos Santos SilvaandRoman Rizzi d792919ddf DEV: Move tokenizers to a gem (#1481)
Also renames the Mixtral tokenizer to Mistral.

See gem at github.com/discourse/discourse_ai-tokenizers


Co-authored-by: Roman Rizzi <[email protected]>
2025-07-02 14:43:03 -03:00
Roman Rizzi 75fb37144f FEATURE: Use personas for generating hypothetical posts (#1482)
* FEATURE: Use personas for generating hypothetica posts

* Update prompt
2025-07-02 10:56:38 -03:00
Sam 40fa527633 FIX: cross talk when in ai helper (#1478)
Previous to this change we reused channels for proofreading progress and
ai helper progress

The new changeset ensures each POST to stream progress gets a dedicated
message bus channel

This fixes a class of issues where the wrong information could be displayed
to end users on subsequent proofreading or helper calls

* fix tests

* fix implementation (got to subscribe at 0)
2025-07-01 18:02:16 +10:00
Yuriy Kurant 8527279594 dev: removes messages section from sidebar (#1477) 2025-07-01 17:23:11 +10:00
Roman Rizzi 5ca7d5f256 FIX: Strip uploads from msg when searching for rag fragments (#1475) 2025-06-30 15:03:17 -03:00
Natalie Tay a94daa14e2 FIX: Return no topics when embeddings is disabled (#1473)
When an invalid model is set for embeddings, topics do not load even if embeddings is disabled.

Error:
## RuntimeError in TopicsController#show
Invalid embeddings selected model

This commit checks for valid settings before attempting to load related topics.
2025-06-30 17:45:04 +08:00
Kris 262bd8b145 UX: add filter to features page, update styles (#1471)
* UX: add filter to features page, update styles

* merge fix

* update toggle spec

* test fix
2025-06-30 09:26:53 +10:00