Component Diagrams

6 components14 relations59 files

Layertestingutility

Details

Static architecture map of 59 files (py, sh) in 6 components: eval: inference, topogpt3: model, topogpt3: inference_hrm, topogpt3: tasks, topogpt3: jlens, eval: data_pipeline. Structure is extracted statically without LLM calls; component prose may be enriched locally.

Relations

eval: inference calls into topogpt3: model — 4 imports, 21 calls across the boundary
app.py::run_training@112 → topogpt3/train.py::TopoGPT3Config · app.py::run_training@115 → topogpt3/train.py::TopoGPT3Trainer · app.py::run_training@117 → topogpt3/train.py::prepare_all
eval: inference calls into topogpt3: inference_hrm — 1 imports, 9 calls across the boundary
app.py::run_inference_hrm@85 → topogpt3/inference_hrm.py::RecursiveReasoningConfig · app.py::run_inference_hrm@90 → topogpt3/inference_hrm.py::HRMInferenceSettings · app.py::run_inference_hrm@101 → topogpt3/inference_hrm.py::HRMInferencePipeline
eval: inference depends on topogpt3: jlens — 2 imports, 0 calls across the boundary
eval: inference calls into eval: data_pipeline — 1 imports, 2 calls across the boundary
eval/harness.py::run_one_test_sandboxed@184 → eval/sandbox.py::SandboxConfig · eval/harness.py::run_one_test_sandboxed@188 → eval/sandbox.py::safe_exec
topogpt3: model calls into eval: inference — 1 imports, 2 calls across the boundary
synthetic_dataset.py::_enqueue_sample@466 → eval/governor.py::put · synthetic_dataset.py::_flush_writer@470 → eval/governor.py::put
topogpt3: model calls into topogpt3: inference_hrm — 2 imports, 3 calls across the boundary
infer_exploitgym.py::load_model@162 → topogpt3/checkpoint_meta.py::config_from_checkpoint · topogpt3/model.py::generate_with_continuation@1266 → topogpt3/continuation.py::is_response_complete · topogpt3/model.py::generate_with_continuation@1269 → topogpt3/continuation.py::extract_tail_for_continuation
topogpt3: inference_hrm calls into topogpt3: model — 3 imports, 30 calls across the boundary
tests/test_structured_output.py::tokenizer@24 → topogpt3/model.py::BPETokenizer · tests/test_structured_output.py::test_json_schema_accepts_and_terminates@118 → topogpt3/model.py::encode · tests/test_structured_output.py::test_regex_masks_illegal_first_token@131 → topogpt3/model.py::encode
topogpt3: inference_hrm calls into topogpt3: tasks — 7 imports, 10 calls across the boundary
topogpt3/api_server.py::list_models@957 → topogpt3/fallback.py::models_payload · topogpt3/api_server.py::_resolve_generation@961 → topogpt3/fallback.py::resolve_profile · topogpt3/api_server.py::completions@1026 → topogpt3/chat.py::_fmt_tool_defs
topogpt3: tasks calls into topogpt3: model — 1 imports, 4 calls across the boundary
eval/tasks.py::build_generate_fn@210 → topogpt3/model.py::BPETokenizer · eval/tasks.py::build_generate_fn@212 → topogpt3/model.py::TopoGPT2 · eval/tasks.py::generate@225 → topogpt3/model.py::encode
topogpt3: tasks calls into topogpt3: inference_hrm — 1 imports, 1 calls across the boundary
eval/tasks.py::build_generate_fn@209 → topogpt3/checkpoint_meta.py::config_from_checkpoint
topogpt3: tasks calls into eval: data_pipeline — 1 imports, 1 calls across the boundary
topogpt3/tools_agent.py::execute_tool@127 → eval/sandbox.py::safe_exec
topogpt3: jlens depends on eval: inference — 1 imports, 0 calls across the boundary
topogpt3: jlens calls into topogpt3: model — 5 imports, 11 calls across the boundary
tests/test_lens_model.py::test_from_topogpt2_config@29 → topogpt3/model.py::TopoGPT2Config · tests/test_lens_model.py::test_forward_output_shape@55 → topogpt3/model.py::model · tests/test_lens_model.py::test_encode_with_tokenizer@101 → topogpt3/model.py::BPETokenizer
topogpt3: jlens depends on topogpt3: inference_hrm — 2 imports, 0 calls across the boundary

eval: inference (1) [Expand]

eval: inference groups 12 files (py) around load_humaneval, build_prompt, extract_candidate, run_one_test, run_one_test_sandboxed. Drop-in entry point that demonstrates how to use the topogpt3 package.

Related Classes/Methods:

eval/harness.py::load_humaneval#L59-L74
eval/harness.py::build_prompt#L75-L99
eval/harness.py::extract_candidate#L100-L149
eval/harness.py::run_one_test#L150-L171
eval/harness.py::run_one_test_sandboxed#L172-L194
eval/harness.py::make_sampler#L195-L203
eval/harness.py::completion_for_problem#L204-L216
eval/harness.py::ModelLoader#L217-L219

Sub-components

eval (1.1)

eval groups 9 files (py) around load_humaneval, build_prompt, extract_candidate, run_one_test, run_one_test_sandboxed. Streaming + governance for autoregressive generation.

root (1.2)

root groups 1 files (py) around run_inference, run_inference_hrm, run_training, _build_parser, main. Drop-in entry point that demonstrates how to use the topogpt3 package.

topogpt3 (1.3)

topogpt3 groups 2 files (py) around ScalePreset, InferenceSettings, scale_presets, preset, validate. TopoGPT3: complex-valued spectral language model for code.

Files

app.py
eval/governor.py
eval/governor_smoke.py
eval/harness.py
eval/integration_smoke.py
eval/noise_sweep.py
eval/repair.py
eval/samplers.py
eval/smoke.py
eval/temp_sweep.py
topogpt3/__init__.py
topogpt3/inference.py

topogpt3: model (2) [Expand]

topogpt3: model groups 12 files (py) around TopoGPT2Config, __post_init__, setup_logger, set_seed, QuaternionOps. Diagnostico estatico de un checkpoint TopoGPT3 congelado.

Related Classes/Methods:

topogpt3/model.py::TopoGPT2Config#L58-L162
topogpt3/model.py::__post_init__#L163-L195
topogpt3/model.py::setup_logger#L196-L205
topogpt3/model.py::set_seed#L206-L217
topogpt3/model.py::QuaternionOps#L218-L225
topogpt3/model.py::hamilton_product#L226-L237
topogpt3/model.py::normalize#L238-L241
topogpt3/model.py::conjugate#L242-L246

Sub-components

eval (2.1)

eval groups 2 files (py) around phase_discretization, synthetic_winding, static_kappa, context_length_diagnostic, main. Diagnostico estatico de un checkpoint TopoGPT3 congelado.

root (2.2)

root groups 3 files (py) around LLMBackend, generate, name, GroqBackend, __init__. Standalone inference script for TopoExploit.

topogpt3 (2.3)

topogpt3 groups 7 files (py) around TopoGPT2Config, __post_init__, setup_logger, set_seed, QuaternionOps. Elastic Weight Consolidation (EWC) and Experience Replay for TopoGPT3.

Files

eval/diag_static.py
eval/hodge_cm_ablation.py
infer_exploitgym.py
synthetic_dataset.py
topogpt3/ewc.py
topogpt3/exploitgym_config.py
topogpt3/exploitgym_loader.py
topogpt3/hodge_cm.py
topogpt3/merged_config.py
topogpt3/model.py
topogpt3/train.py
transfer_weights.py

topogpt3: inference_hrm (3) [Expand]

topogpt3: inference_hrm groups 8 files (py) around _count_unclosed_brackets, _count_unclosed_fences, is_response_complete, extract_tail_for_continuation, split_at_last_newline. Checkpoint architecture discovery tests.

Related Classes/Methods:

topogpt3/continuation.py::_count_unclosed_brackets#L25-L35
topogpt3/continuation.py::_count_unclosed_fences#L36-L44
topogpt3/continuation.py::is_response_complete#L45-L74
topogpt3/continuation.py::extract_tail_for_continuation#L75-L104
topogpt3/continuation.py::split_at_last_newline#L105-L114
topogpt3/checkpoint_meta.py::read_state#L24-L34
topogpt3/checkpoint_meta.py::probe_checkpoint#L35-L80
topogpt3/checkpoint_meta.py::config_from_checkpoint#L81-L119

Sub-components

tests (3.1)

tests groups 2 files (py) around test_read_state_missing, test_probe_t3_small, test_probe_exploitgym_meta, test_config_matches_weights, tokenizer. Checkpoint architecture discovery tests.

topogpt3 (3.2)

topogpt3 groups 6 files (py) around _count_unclosed_brackets, _count_unclosed_fences, is_response_complete, extract_tail_for_continuation, split_at_last_newline. OpenAI-compatible HTTP API server so TopoGPT3 can be used as a backend for coding...

Files

tests/test_checkpoint_meta.py
tests/test_structured_output.py
topogpt3/api_server.py
topogpt3/checkpoint_meta.py
topogpt3/continuation.py
topogpt3/inference_hrm.py
topogpt3/structured_output.py
topogpt3/tokenizer_tools.py

topogpt3: tasks (4) [Expand]

topogpt3: tasks groups 7 files (py) around pre_processing_chat, post_processing_chat, _fmt_tool_defs, _tool_calls_text, apply_chat_template. Generic task eval: JSONL cases + safe verifiers.

Related Classes/Methods:

topogpt3/chat.py::pre_processing_chat#L44-L57
topogpt3/chat.py::post_processing_chat#L58-L63
topogpt3/chat.py::_fmt_tool_defs#L64-L88
topogpt3/chat.py::_tool_calls_text#L89-L115
topogpt3/chat.py::apply_chat_template#L116-L167
topogpt3/chat.py::parse_tool_calls#L168-L177
topogpt3/chat.py::parse_thinking#L178-L184
topogpt3/chat.py::split_reasoning_content#L185-L196

Sub-components

eval (4.1)

eval groups 1 files (py) around _extract_code, _extract_json, verify_tool_call, verify_json, verify_regex. Generic task eval: JSONL cases + safe verifiers.

tests (4.2)

tests groups 2 files (py) around _gen, test_apt_chain_cases_full_pass, test_regex_verifier, test_tool_call_verifier_rejects_wrong_target, _call. Generic task eval tests (no model load).

topogpt3 (4.3)

topogpt3 groups 4 files (py) around pre_processing_chat, post_processing_chat, _fmt_tool_defs, _tool_calls_text, apply_chat_template. Chat template + special tokens for TopoGPT3.

Files

eval/tasks.py
tests/test_eval_tasks.py
tests/test_fallback.py
topogpt3/chat.py
topogpt3/eval_toolcall.py
topogpt3/fallback.py
topogpt3/tools_agent.py

topogpt3: jlens (5)

topogpt3: jlens groups 5 files (py) around LensModel, encode, forward, unembed, TopoGPT3LensConfig. TestValidPositionMask: Feature: valid_position_mask excludes attention-sink and final...

Related Classes/Methods:

topogpt3/lens_model.py::LensModel#L23-L39
topogpt3/lens_model.py::encode#L40-L44
topogpt3/lens_model.py::forward#L45-L51
topogpt3/lens_model.py::unembed#L52-L58
topogpt3/lens_model.py::TopoGPT3LensConfig#L59-L83
topogpt3/lens_model.py::from_topogpt2_config#L84-L103
topogpt3/lens_model.py::probe_checkpoint#L104-L141
topogpt3/lens_model.py::_TopoGPT3ResidualForward#L142-L149

Files

tests/test_jlens.py
tests/test_lens_model.py
topogpt3/__main__.py
topogpt3/jlens.py
topogpt3/lens_model.py

eval: data_pipeline (6)

eval: data_pipeline groups 4 files (py) around SandboxConfig, _names_imported, _blocked_dunder_access, _max_depth, d. Sandbox for executing model-generated code during HumanEval evaluation.

Related Classes/Methods:

eval/sandbox.py::SandboxConfig#L53-L99
eval/sandbox.py::_names_imported#L100-L113
eval/sandbox.py::_blocked_dunder_access#L114-L122
eval/sandbox.py::_max_depth#L123-L124
eval/sandbox.py::d#L125-L132
eval/sandbox.py::check_safety#L133-L253
eval/sandbox.py::_build_worker_src#L254-L269
eval/sandbox.py::safe_exec#L270-L372

Files

eval/sandbox.py
eval/sandbox_smoke.py
tests/test_data_pipeline.py
topogpt3/data_pipeline.py