ruby_llm-contract 0.5.2 → 0.6.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- checksums.yaml +4 -4
- data/CHANGELOG.md +49 -2
- data/Gemfile.lock +2 -2
- data/README.md +173 -134
- data/lib/ruby_llm/contract/concerns/eval_host.rb +25 -6
- data/lib/ruby_llm/contract/eval/model_comparison.rb +33 -11
- data/lib/ruby_llm/contract/eval/recommendation.rb +48 -0
- data/lib/ruby_llm/contract/eval/recommender.rb +132 -0
- data/lib/ruby_llm/contract/eval/report.rb +2 -2
- data/lib/ruby_llm/contract/eval/report_stats.rb +6 -0
- data/lib/ruby_llm/contract/eval/report_storage.rb +18 -12
- data/lib/ruby_llm/contract/eval/retry_optimizer.rb +221 -0
- data/lib/ruby_llm/contract/eval.rb +3 -0
- data/lib/ruby_llm/contract/rake_task.rb +83 -0
- data/lib/ruby_llm/contract/step/base.rb +30 -0
- data/lib/ruby_llm/contract/step/retry_executor.rb +9 -3
- data/lib/ruby_llm/contract/step/retry_policy.rb +27 -14
- data/lib/ruby_llm/contract/version.rb +1 -1
- data/lib/ruby_llm/contract.rb +21 -0
- metadata +4 -1
|
@@ -9,13 +9,13 @@ module RubyLLM
|
|
|
9
9
|
DEFAULT_RETRY_ON = %i[validation_failed parse_error adapter_error].freeze
|
|
10
10
|
|
|
11
11
|
def initialize(models: nil, attempts: nil, retry_on: nil, &block)
|
|
12
|
-
@
|
|
12
|
+
@configs = []
|
|
13
13
|
@retryable_statuses = DEFAULT_RETRY_ON.dup
|
|
14
14
|
|
|
15
15
|
if block
|
|
16
16
|
@max_attempts = 1
|
|
17
17
|
instance_eval(&block)
|
|
18
|
-
warn_no_retry! if @max_attempts == 1 && @
|
|
18
|
+
warn_no_retry! if @max_attempts == 1 && @configs.empty?
|
|
19
19
|
else
|
|
20
20
|
apply_keywords(models: models, attempts: attempts, retry_on: retry_on)
|
|
21
21
|
end
|
|
@@ -28,14 +28,18 @@ module RubyLLM
|
|
|
28
28
|
validate_max_attempts!
|
|
29
29
|
end
|
|
30
30
|
|
|
31
|
-
def escalate(*
|
|
32
|
-
@
|
|
33
|
-
@max_attempts = @
|
|
31
|
+
def escalate(*config_list)
|
|
32
|
+
@configs = config_list.flatten.map { |c| normalize_config(c).freeze }.freeze
|
|
33
|
+
@max_attempts = @configs.length if @max_attempts < @configs.length
|
|
34
34
|
end
|
|
35
35
|
alias models escalate
|
|
36
36
|
|
|
37
37
|
def model_list
|
|
38
|
-
@
|
|
38
|
+
@configs.map { |c| c[:model] }.freeze
|
|
39
|
+
end
|
|
40
|
+
|
|
41
|
+
def config_list
|
|
42
|
+
@configs
|
|
39
43
|
end
|
|
40
44
|
|
|
41
45
|
def retry_on(*statuses)
|
|
@@ -46,29 +50,38 @@ module RubyLLM
|
|
|
46
50
|
retryable_statuses.include?(result.status)
|
|
47
51
|
end
|
|
48
52
|
|
|
49
|
-
def
|
|
50
|
-
if @
|
|
51
|
-
@
|
|
53
|
+
def config_for_attempt(attempt, default_config)
|
|
54
|
+
if @configs.any?
|
|
55
|
+
@configs[attempt] || @configs.last
|
|
52
56
|
else
|
|
53
|
-
|
|
57
|
+
default_config
|
|
54
58
|
end
|
|
55
59
|
end
|
|
56
60
|
|
|
61
|
+
def model_for_attempt(attempt, default_model)
|
|
62
|
+
config_for_attempt(attempt, { model: default_model })[:model]
|
|
63
|
+
end
|
|
64
|
+
|
|
57
65
|
private
|
|
58
66
|
|
|
59
67
|
def apply_keywords(models:, attempts:, retry_on:)
|
|
60
68
|
if models
|
|
61
|
-
@
|
|
62
|
-
@max_attempts = @
|
|
69
|
+
@configs = Array(models).map { |m| normalize_config(m).freeze }.freeze
|
|
70
|
+
@max_attempts = @configs.length
|
|
63
71
|
else
|
|
64
72
|
@max_attempts = attempts || 1
|
|
65
73
|
end
|
|
66
74
|
@retryable_statuses = Array(retry_on).dup if retry_on
|
|
67
75
|
end
|
|
68
76
|
|
|
77
|
+
def normalize_config(entry)
|
|
78
|
+
RubyLLM::Contract.normalize_candidate_config(entry)
|
|
79
|
+
end
|
|
80
|
+
|
|
69
81
|
def warn_no_retry!
|
|
70
|
-
warn "[ruby_llm-contract] retry_policy has max_attempts=1 with no
|
|
71
|
-
"This means no actual retry will happen. Add `attempts 2` or
|
|
82
|
+
warn "[ruby_llm-contract] retry_policy has max_attempts=1 with no configs. " \
|
|
83
|
+
"This means no actual retry will happen. Add `attempts 2` or " \
|
|
84
|
+
'`escalate "model1", "model2"`.'
|
|
72
85
|
end
|
|
73
86
|
|
|
74
87
|
def validate_max_attempts!
|
data/lib/ruby_llm/contract.rb
CHANGED
|
@@ -78,6 +78,27 @@ module RubyLLM
|
|
|
78
78
|
Thread.current[:ruby_llm_contract_reloading] = false
|
|
79
79
|
end
|
|
80
80
|
|
|
81
|
+
def normalize_candidate_config(entry)
|
|
82
|
+
case entry
|
|
83
|
+
when String
|
|
84
|
+
raise ArgumentError, "Candidate model must be a non-empty String" if entry.strip.empty?
|
|
85
|
+
|
|
86
|
+
{ model: entry.strip }
|
|
87
|
+
when Hash
|
|
88
|
+
model = entry[:model] || entry["model"]
|
|
89
|
+
unless model.is_a?(String) && !model.strip.empty?
|
|
90
|
+
raise ArgumentError, "Candidate config must include a non-empty String :model"
|
|
91
|
+
end
|
|
92
|
+
|
|
93
|
+
normalized = { model: model.strip }
|
|
94
|
+
effort = entry[:reasoning_effort] || entry["reasoning_effort"]
|
|
95
|
+
normalized[:reasoning_effort] = effort if effort
|
|
96
|
+
normalized
|
|
97
|
+
else
|
|
98
|
+
raise ArgumentError, "Expected String or Hash, got #{entry.class}"
|
|
99
|
+
end
|
|
100
|
+
end
|
|
101
|
+
|
|
81
102
|
private
|
|
82
103
|
|
|
83
104
|
# Filter stale hosts, deduplicate by name (last wins), prune registry in-place
|
metadata
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
--- !ruby/object:Gem::Specification
|
|
2
2
|
name: ruby_llm-contract
|
|
3
3
|
version: !ruby/object:Gem::Version
|
|
4
|
-
version: 0.
|
|
4
|
+
version: 0.6.2
|
|
5
5
|
platform: ruby
|
|
6
6
|
authors:
|
|
7
7
|
- Justyna
|
|
@@ -130,10 +130,13 @@ files:
|
|
|
130
130
|
- lib/ruby_llm/contract/eval/prompt_diff_comparator.rb
|
|
131
131
|
- lib/ruby_llm/contract/eval/prompt_diff_presenter.rb
|
|
132
132
|
- lib/ruby_llm/contract/eval/prompt_diff_serializer.rb
|
|
133
|
+
- lib/ruby_llm/contract/eval/recommendation.rb
|
|
134
|
+
- lib/ruby_llm/contract/eval/recommender.rb
|
|
133
135
|
- lib/ruby_llm/contract/eval/report.rb
|
|
134
136
|
- lib/ruby_llm/contract/eval/report_presenter.rb
|
|
135
137
|
- lib/ruby_llm/contract/eval/report_stats.rb
|
|
136
138
|
- lib/ruby_llm/contract/eval/report_storage.rb
|
|
139
|
+
- lib/ruby_llm/contract/eval/retry_optimizer.rb
|
|
137
140
|
- lib/ruby_llm/contract/eval/runner.rb
|
|
138
141
|
- lib/ruby_llm/contract/eval/step_expectation_applier.rb
|
|
139
142
|
- lib/ruby_llm/contract/eval/step_result_normalizer.rb
|