Text Generation
Transformers
Safetensors
PEFT
English
Chinese
qwen3_5
image-text-to-text
veriloop
veriloop-coder
code
coding-agent
software-engineering
repository-understanding
tool-use
lora
harness-engineering
evidence-binding
rollback
uncertainty-calibration
long-context
open-weights
conversational
Instructions to use veriloop-lab/veriloop-coder-e1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use veriloop-lab/veriloop-coder-e1 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="veriloop-lab/veriloop-coder-e1") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] pipe(text=messages)# Load model directly from transformers import AutoProcessor, AutoModelForImageTextToText processor = AutoProcessor.from_pretrained("veriloop-lab/veriloop-coder-e1") model = AutoModelForImageTextToText.from_pretrained("veriloop-lab/veriloop-coder-e1") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] inputs = processor.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) outputs = model.generate(**inputs, max_new_tokens=40) print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - PEFT
How to use veriloop-lab/veriloop-coder-e1 with PEFT:
Task type is invalid.
- Notebooks
- Google Colab
- Kaggle
- Local Apps
- vLLM
How to use veriloop-lab/veriloop-coder-e1 with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "veriloop-lab/veriloop-coder-e1" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "veriloop-lab/veriloop-coder-e1", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/veriloop-lab/veriloop-coder-e1
- SGLang
How to use veriloop-lab/veriloop-coder-e1 with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "veriloop-lab/veriloop-coder-e1" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "veriloop-lab/veriloop-coder-e1", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "veriloop-lab/veriloop-coder-e1" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "veriloop-lab/veriloop-coder-e1", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use veriloop-lab/veriloop-coder-e1 with Docker Model Runner:
docker model run hf.co/veriloop-lab/veriloop-coder-e1
Upload model/toolspec_adapter_plan.json with huggingface_hub
Browse files- model/toolspec_adapter_plan.json +218 -0
model/toolspec_adapter_plan.json
ADDED
|
@@ -0,0 +1,218 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"backbone": "/public/wang_libo/veriloop_coder_e1/model",
|
| 3 |
+
"dataset_summary": {
|
| 4 |
+
"eval_size": 25,
|
| 5 |
+
"modes": [
|
| 6 |
+
"argument_shaping",
|
| 7 |
+
"harness_envelope",
|
| 8 |
+
"precondition_gating",
|
| 9 |
+
"receipt_and_validation",
|
| 10 |
+
"reverse_engineering",
|
| 11 |
+
"runtime_session",
|
| 12 |
+
"self_check_loop",
|
| 13 |
+
"tool_trigger",
|
| 14 |
+
"worktree_permission"
|
| 15 |
+
],
|
| 16 |
+
"tools": [
|
| 17 |
+
"apply_patch_bundle",
|
| 18 |
+
"browser_search",
|
| 19 |
+
"repo_reverse_engineer",
|
| 20 |
+
"run_ci_validation",
|
| 21 |
+
"self_check_repair_loop"
|
| 22 |
+
],
|
| 23 |
+
"train_size": 70
|
| 24 |
+
},
|
| 25 |
+
"excluded_surfaces": [
|
| 26 |
+
"(^|\\.)lm_head($|\\.)::Do not retune final token head; too broad and evaluation-heavy.",
|
| 27 |
+
"(^|\\.)embed_tokens($|\\.)::Embedding surgery risks broad semantic drift.",
|
| 28 |
+
"(^|\\.)norm($|\\.)::Global norm tuning can destabilize calibration across scenes.",
|
| 29 |
+
"attnres|attention_residual::Block AttnRes may be mounted structurally but is never a PEFT target.",
|
| 30 |
+
"dualpath::DualPath is serving/runtime infrastructure only.",
|
| 31 |
+
"mhc|hyper[-_]?connection::mHC-inspired stability hooks remain structural, not PEFT surfaces.",
|
| 32 |
+
"rope|rotary::RoPE/context surgery is handled architecturally, not by narrow PEFT here.",
|
| 33 |
+
"kvcache|kv_cache::KV-cache runtime surfaces are not PEFT targets.",
|
| 34 |
+
"(^|\\.)memory(_store|_bank)?($|\\.)::Persistent memory stores are harness/runtime policy surfaces, not PEFT targets."
|
| 35 |
+
],
|
| 36 |
+
"notes": [
|
| 37 |
+
"Primary route is host-surface-first toolspec probe training, not broad coding SFT.",
|
| 38 |
+
"This adapter is allowed to shape runtime protocol obedience, validator alignment, self-check willingness, and reverse-engineering readiness.",
|
| 39 |
+
"Backbone attention, MoE router/expert weights, DualPath, Block AttnRes, and mHC remain excluded.",
|
| 40 |
+
"Explicit host/runtime bridge leaves default to LoRA-narrow attachment because IA3 feedforward constraints do not fit this custom side-car surface design."
|
| 41 |
+
],
|
| 42 |
+
"peft_method": "lora_narrow",
|
| 43 |
+
"product_line": "veriloop_coder",
|
| 44 |
+
"recipe": {
|
| 45 |
+
"adapter_family": "runtime_harness",
|
| 46 |
+
"backbone": "/public/wang_libo/veriloop_coder_e1/model",
|
| 47 |
+
"backbone_family": "qwen_dense",
|
| 48 |
+
"excluded_patterns": [
|
| 49 |
+
"(?i)\\bdualpath\\b",
|
| 50 |
+
"(?i)\\bmhc\\b",
|
| 51 |
+
"(?i)\\bfull[_\\- ]?attnres\\b",
|
| 52 |
+
"(?i)\\battnres(_full)?\\b",
|
| 53 |
+
"(?i)\\brouter\\b",
|
| 54 |
+
"(?i)\\bexperts?\\b",
|
| 55 |
+
"(?i)\\bmoe\\b.*\\b(gate|router|expert)\\b",
|
| 56 |
+
"(?i)\\brope\\b.*\\b(freq|inv_freq|theta|rotary)\\b",
|
| 57 |
+
"(?i)\\bkvcache\\b",
|
| 58 |
+
"(?i)\\bposition_embedding\\b",
|
| 59 |
+
"(?i)\\bembed(tokens|ding)?\\b",
|
| 60 |
+
"(?i)\\blm_head\\b"
|
| 61 |
+
],
|
| 62 |
+
"harness_constraints": [
|
| 63 |
+
"Harness Engineering remains the primary convergence layer.",
|
| 64 |
+
"Adapter must not bypass runtime orchestrator / validator / rollback loops.",
|
| 65 |
+
"Adapter outputs remain subordinate to VeriLoop control-plane decisions.",
|
| 66 |
+
"Adapter must not create hidden prompt-style memory authority.",
|
| 67 |
+
"Adapter must improve runtime protocol obedience, not free-form style drift.",
|
| 68 |
+
"Tool legality, permission discipline, session continuity, and worktree hygiene must remain first-class."
|
| 69 |
+
],
|
| 70 |
+
"hyperparams": {
|
| 71 |
+
"alpha": 0,
|
| 72 |
+
"bias": "none",
|
| 73 |
+
"dropout": 0.0,
|
| 74 |
+
"fan_in_fan_out": false,
|
| 75 |
+
"modules_to_save": [],
|
| 76 |
+
"r": 0,
|
| 77 |
+
"task_type": "CAUSAL_LM"
|
| 78 |
+
},
|
| 79 |
+
"merge_policy": "side_load",
|
| 80 |
+
"metadata": {
|
| 81 |
+
"allow_backbone_bridge": false,
|
| 82 |
+
"allow_vla_action_expert": false,
|
| 83 |
+
"harness_first": true,
|
| 84 |
+
"prefer_explicit_heads": true,
|
| 85 |
+
"prefer_qlora_for_backbone_bridge": true,
|
| 86 |
+
"require_harness_first": true,
|
| 87 |
+
"selector_group_count": 2,
|
| 88 |
+
"strict_narrow_scope": true,
|
| 89 |
+
"toolspec_probe_training": true,
|
| 90 |
+
"trainer": "veriloop.toolspec_adapter_trainer.v5.qwen36"
|
| 91 |
+
},
|
| 92 |
+
"notes": [
|
| 93 |
+
"Harness Engineering is primary; PEFT is limited to obedience-facing, interface-facing support surfaces.",
|
| 94 |
+
"Backbone bridge tuning disabled explicitly; selector stays on custom surfaces or no-op.",
|
| 95 |
+
"Backbone family inferred as qwen_dense.",
|
| 96 |
+
"PEFT method resolved as ia3_head_only.",
|
| 97 |
+
"Recipe is harness-first: runtime convergence remains in VeriLoop control-plane + harness, not in broad weight surgery.",
|
| 98 |
+
"Block AttnRes, DualPath, mHC hooks, RoPE, KV-cache, and broad MoE routing remain structurally excluded."
|
| 99 |
+
],
|
| 100 |
+
"peft_method": "lora_narrow",
|
| 101 |
+
"precision_policy": "auto",
|
| 102 |
+
"product_line": "veriloop_coder",
|
| 103 |
+
"regression_requirements": [
|
| 104 |
+
"Must pass PEFT regression guard structural policy checks.",
|
| 105 |
+
"Must not introduce forbidden backbone/serving structural targets.",
|
| 106 |
+
"Must preserve harness regression envelope for the selected product line.",
|
| 107 |
+
"Runtime protocol obedience must improve or hold.",
|
| 108 |
+
"Tool trigger accuracy, permission discipline, and worktree continuity must not regress."
|
| 109 |
+
],
|
| 110 |
+
"target_groups": [
|
| 111 |
+
{
|
| 112 |
+
"alpha": 16,
|
| 113 |
+
"dropout": 0.05,
|
| 114 |
+
"name": "group_1_custom_runtime_harness_bridge",
|
| 115 |
+
"rank": 8,
|
| 116 |
+
"rationale": "Runtime / harness obedience should attach to explicit interface bridges before any backbone fallback.",
|
| 117 |
+
"surface": "custom_runtime_harness_bridge",
|
| 118 |
+
"target_modules": [
|
| 119 |
+
"failure_signal_bridge.rollback_bridge",
|
| 120 |
+
"request_normalizer",
|
| 121 |
+
"request_normalizer.adapter",
|
| 122 |
+
"rollback_adapter",
|
| 123 |
+
"rollback_adapter.head",
|
| 124 |
+
"runtime_harness_adapter",
|
| 125 |
+
"runtime_harness_adapter.bridge",
|
| 126 |
+
"tool_protocol_adapter",
|
| 127 |
+
"tool_protocol_adapter.bridge",
|
| 128 |
+
"toolspec_bridge",
|
| 129 |
+
"toolspec_bridge.adapter",
|
| 130 |
+
"toolspec_head",
|
| 131 |
+
"toolspec_head.param_schema_adapter",
|
| 132 |
+
"toolspec_head.postcondition_adapter",
|
| 133 |
+
"toolspec_head.precondition_adapter",
|
| 134 |
+
"toolspec_head.receipt_formatter",
|
| 135 |
+
"toolspec_head.trigger_gate",
|
| 136 |
+
"toolspec_head.validator_gate",
|
| 137 |
+
"validator_feedback_loop.rollback_adapter"
|
| 138 |
+
]
|
| 139 |
+
},
|
| 140 |
+
{
|
| 141 |
+
"alpha": 8,
|
| 142 |
+
"dropout": 0.0,
|
| 143 |
+
"name": "group_2_custom_memory_boundary_bridge",
|
| 144 |
+
"rank": 4,
|
| 145 |
+
"rationale": "Session continuity should bind to boundary-aware memory packet surfaces rather than broad backbone tuning.",
|
| 146 |
+
"surface": "custom_memory_boundary_bridge",
|
| 147 |
+
"target_modules": [
|
| 148 |
+
"episodic_memory",
|
| 149 |
+
"episodic_memory.adapter",
|
| 150 |
+
"memory_boundary_guard",
|
| 151 |
+
"memory_boundary_guard.adapter",
|
| 152 |
+
"memory_boundary_guard.rollback_filter",
|
| 153 |
+
"session_compactor",
|
| 154 |
+
"session_compactor.adapter"
|
| 155 |
+
]
|
| 156 |
+
}
|
| 157 |
+
],
|
| 158 |
+
"target_modules": [
|
| 159 |
+
"failure_signal_bridge.rollback_bridge",
|
| 160 |
+
"request_normalizer",
|
| 161 |
+
"request_normalizer.adapter",
|
| 162 |
+
"rollback_adapter",
|
| 163 |
+
"rollback_adapter.head",
|
| 164 |
+
"runtime_harness_adapter",
|
| 165 |
+
"runtime_harness_adapter.bridge",
|
| 166 |
+
"tool_protocol_adapter",
|
| 167 |
+
"tool_protocol_adapter.bridge",
|
| 168 |
+
"toolspec_bridge",
|
| 169 |
+
"toolspec_bridge.adapter",
|
| 170 |
+
"toolspec_head",
|
| 171 |
+
"toolspec_head.param_schema_adapter",
|
| 172 |
+
"toolspec_head.postcondition_adapter",
|
| 173 |
+
"toolspec_head.precondition_adapter",
|
| 174 |
+
"toolspec_head.receipt_formatter",
|
| 175 |
+
"toolspec_head.trigger_gate",
|
| 176 |
+
"toolspec_head.validator_gate",
|
| 177 |
+
"validator_feedback_loop.rollback_adapter",
|
| 178 |
+
"episodic_memory",
|
| 179 |
+
"episodic_memory.adapter",
|
| 180 |
+
"memory_boundary_guard",
|
| 181 |
+
"memory_boundary_guard.adapter",
|
| 182 |
+
"memory_boundary_guard.rollback_filter",
|
| 183 |
+
"session_compactor",
|
| 184 |
+
"session_compactor.adapter"
|
| 185 |
+
],
|
| 186 |
+
"version": "veriloop.lora_recipe_veriloop.v2"
|
| 187 |
+
},
|
| 188 |
+
"selected_surfaces": [
|
| 189 |
+
"custom_runtime_harness_bridge",
|
| 190 |
+
"custom_validator_bridge",
|
| 191 |
+
"custom_memory_boundary_bridge"
|
| 192 |
+
],
|
| 193 |
+
"selected_target_modules": [
|
| 194 |
+
"request_normalizer",
|
| 195 |
+
"request_normalizer.adapter",
|
| 196 |
+
"runtime_harness_adapter",
|
| 197 |
+
"runtime_harness_adapter.bridge",
|
| 198 |
+
"tool_protocol_adapter",
|
| 199 |
+
"tool_protocol_adapter.bridge",
|
| 200 |
+
"toolspec_bridge",
|
| 201 |
+
"toolspec_bridge.adapter",
|
| 202 |
+
"toolspec_head",
|
| 203 |
+
"toolspec_head.param_schema_adapter",
|
| 204 |
+
"toolspec_head.postcondition_adapter",
|
| 205 |
+
"toolspec_head.precondition_adapter",
|
| 206 |
+
"toolspec_head.receipt_formatter",
|
| 207 |
+
"toolspec_head.trigger_gate",
|
| 208 |
+
"toolspec_head.validator_gate",
|
| 209 |
+
"validator_feedback_bridge",
|
| 210 |
+
"validator_feedback_bridge.adapter"
|
| 211 |
+
],
|
| 212 |
+
"selection_mode": "minimal",
|
| 213 |
+
"version": "veriloop.toolspec_adapter_trainer.v5.qwen36",
|
| 214 |
+
"warnings": [
|
| 215 |
+
"Harness Engineering is primary; PEFT is limited to obedience-facing, interface-facing support surfaces.",
|
| 216 |
+
"Backbone bridge tuning disabled explicitly; selector stays on custom surfaces or no-op."
|
| 217 |
+
]
|
| 218 |
+
}
|