Small llms improvements (#400)

* Initial implementation

* FIX keys

* Add langfuse evals support

* FIX trace upload

* Delete .claude/settings.local.json

Signed-off-by: soky srm <sokysrm@gmail.com>

* Update client.rb

* Small LLMs improvements

* Keep batch size normal

* Update categorizer

* FIX json mode

* Add reasonable alternative to matching

* FIX thinking blocks for llms

* Implement json mode support with AUTO mode

* Make auto default for everyone

* FIX linter

* Address review

* Allow export manual categories

* FIX user export

* FIX oneshot example pollution

* Update categorization_golden_v1.yml

* Update categorization_golden_v1.yml

* Trim to 100 items

* Update auto_categorizer.rb

* FIX for auto retry in auto mode

* Separate the Eval Logic from the Auto-Categorizer

The expected_null_count parameter conflates eval-specific logic with production categorization logic.

* Force json mode on evals

* Introduce a more mixed dataset

150 items, performance from a local model:

By Difficulty:
  easy: 93.22% accuracy (55/59)
  medium: 93.33% accuracy (42/45)
  hard: 92.86% accuracy (26/28)
  edge_case: 100.0% accuracy (18/18)

* Improve datasets

Remove Data leakage from prompts

* Create eval runs as "pending"

---------

Signed-off-by: soky srm <sokysrm@gmail.com>
Signed-off-by: Juan José Mata <juanjo.mata@gmail.com>
Co-authored-by: Juan José Mata <juanjo.mata@gmail.com>
This commit is contained in:
soky srm
2025-12-07 18:11:34 +01:00
committed by GitHub
parent bf90cad9a0
commit 88952e4714
34 changed files with 11027 additions and 42 deletions

View File

@@ -51,7 +51,7 @@ class Provider::Openai < Provider
@uri_base.present?
end
def auto_categorize(transactions: [], user_categories: [], model: "", family: nil)
def auto_categorize(transactions: [], user_categories: [], model: "", family: nil, json_mode: nil)
with_provider_response do
raise Error, "Too many transactions to auto-categorize. Max is 25 per request." if transactions.size > 25
if user_categories.blank?
@@ -74,7 +74,8 @@ class Provider::Openai < Provider
user_categories: user_categories,
custom_provider: custom_provider?,
langfuse_trace: trace,
family: family
family: family,
json_mode: json_mode
).auto_categorize
trace&.update(output: result.map(&:to_h))
@@ -83,7 +84,7 @@ class Provider::Openai < Provider
end
end
def auto_detect_merchants(transactions: [], user_merchants: [], model: "", family: nil)
def auto_detect_merchants(transactions: [], user_merchants: [], model: "", family: nil, json_mode: nil)
with_provider_response do
raise Error, "Too many transactions to auto-detect merchants. Max is 25 per request." if transactions.size > 25
@@ -101,7 +102,8 @@ class Provider::Openai < Provider
user_merchants: user_merchants,
custom_provider: custom_provider?,
langfuse_trace: trace,
family: family
family: family,
json_mode: json_mode
).auto_detect_merchants
trace&.update(output: result.map(&:to_h))