Documentation
¶
Index ¶
- type LatencyProfile
- type MockLLM
- func (m *MockLLM) ChatCompletion(ctx context.Context, req llm.CompletionRequest, ...) (*llm.TextStreamResult, error)
- func (m *MockLLM) ChatCompletionNoStream(ctx context.Context, req llm.CompletionRequest, ...) (string, error)
- func (m *MockLLM) CountTokens(_ context.Context, request llm.CompletionRequest, _ ...llm.LanguageModelOption) (int, error)
- func (m *MockLLM) InputTokenLimit() int
- func (m *MockLLM) OutputTokenLimit() int
- type MockProfile
- type ToolArgumentProfile
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
This section is empty.
Types ¶
type LatencyProfile ¶
type LatencyProfile struct {
TTFTMs [2]int `json:"ttft_ms"`
ChunkCount [2]int `json:"chunk_count"`
ChunkIntervalMs [2]int `json:"chunk_interval_ms"`
TotalWallTimeMsPerRequest [2]int `json:"total_wall_time_ms_per_request"`
}
LatencyProfile describes one named latency mix for mock streaming.
type MockLLM ¶
type MockLLM struct {
// contains filtered or unexported fields
}
MockLLM is an in-process llm.LanguageModel for load tests.
func NewMockLLM ¶
func NewMockLLM(profile MockProfile) *MockLLM
NewMockLLM constructs a mock LLM using the given profile (must be validated).
func (*MockLLM) ChatCompletion ¶
func (m *MockLLM) ChatCompletion(ctx context.Context, req llm.CompletionRequest, opts ...llm.LanguageModelOption) (*llm.TextStreamResult, error)
ChatCompletion streams a synthetic response.
func (*MockLLM) ChatCompletionNoStream ¶
func (m *MockLLM) ChatCompletionNoStream(ctx context.Context, req llm.CompletionRequest, opts ...llm.LanguageModelOption) (string, error)
ChatCompletionNoStream returns the final mock text after honoring wall-clock pacing.
func (*MockLLM) CountTokens ¶
func (m *MockLLM) CountTokens(_ context.Context, request llm.CompletionRequest, _ ...llm.LanguageModelOption) (int, error)
CountTokens approximates input tokens across the request's posts. The mock always returns a count, so callers never fall back to EstimateTokens.
func (*MockLLM) InputTokenLimit ¶
InputTokenLimit returns a generous limit for load testing.
func (*MockLLM) OutputTokenLimit ¶
OutputTokenLimit returns a generous limit for load testing.
type MockProfile ¶
type MockProfile struct {
Name string `json:"name"`
Seed int64 `json:"seed"`
LatencyProfiles map[string]LatencyProfile `json:"latency_profiles"`
ProfileWeights map[string]float64 `json:"profile_weights"`
ReasoningSkipProbability float64 `json:"reasoning_skip_probability"`
StreamingEnabled bool `json:"streaming_enabled"`
ToolUseProbability float64 `json:"tool_use_probability"`
ToolWeights map[string]float64 `json:"tool_weights"`
MaxToolRounds int `json:"max_tool_rounds"`
ToolArgumentProfiles map[string]ToolArgumentProfile `json:"tool_argument_profiles,omitempty"`
FinalResponseTemplates []string `json:"final_response_templates"`
}
MockProfile configures load-test LLM behavior.
func DefaultReadSearchHeavyProfile ¶
func DefaultReadSearchHeavyProfile() MockProfile
DefaultReadSearchHeavyProfile returns the documented empirical defaults for read/search-heavy load tests.
func ParseProfile ¶
func ParseProfile(raw json.RawMessage) (MockProfile, error)
ParseProfile merges operator JSON on top of the default profile. Nil, empty, or whitespace-only raw returns the default.
func (MockProfile) Summary ¶
func (p MockProfile) Summary() string
Summary returns a compact operator-facing description for logging.
func (MockProfile) Validate ¶
func (p MockProfile) Validate() error
Validate checks profile invariants.
type ToolArgumentProfile ¶
type ToolArgumentProfile struct {
PostLimits []int `json:"post_limits,omitempty"`
SearchQueries []string `json:"search_queries,omitempty"`
SearchLimits []int `json:"search_limits,omitempty"`
MessageLengths []int `json:"message_lengths,omitempty"`
Usernames []string `json:"usernames,omitempty"`
ChannelIDs []string `json:"channel_ids,omitempty"`
ChannelNames []string `json:"channel_names,omitempty"`
TeamIDs []string `json:"team_ids,omitempty"`
TeamNames []string `json:"team_names,omitempty"`
PostIDs []string `json:"post_ids,omitempty"`
}
ToolArgumentProfile holds optional discrete values for argument generation per tool.
Source Files