Text Generation
Transformers
Safetensors
olmo3
code
reasoning
lora-merged
livecodebench
conversational
Instructions to use modrill/code-think-o7b-20260908 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use modrill/code-think-o7b-20260908 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="modrill/code-think-o7b-20260908") messages = [ {"role": "user", "content": "Who are you?"}, ] pipe(messages)# Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("modrill/code-think-o7b-20260908") model = AutoModelForCausalLM.from_pretrained("modrill/code-think-o7b-20260908", device_map="auto") messages = [ {"role": "user", "content": "Who are you?"}, ] inputs = tokenizer.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(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use modrill/code-think-o7b-20260908 with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "modrill/code-think-o7b-20260908" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "modrill/code-think-o7b-20260908", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/modrill/code-think-o7b-20260908
- SGLang
How to use modrill/code-think-o7b-20260908 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 "modrill/code-think-o7b-20260908" \ --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": "modrill/code-think-o7b-20260908", "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 "modrill/code-think-o7b-20260908" \ --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": "modrill/code-think-o7b-20260908", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use modrill/code-think-o7b-20260908 with Docker Model Runner:
docker model run hf.co/modrill/code-think-o7b-20260908
Add provenance/TRAINING_CONFIG.json
Browse files- provenance/TRAINING_CONFIG.json +187 -0
provenance/TRAINING_CONFIG.json
ADDED
|
@@ -0,0 +1,187 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"schema": "code-sixarm-training-config/v4",
|
| 3 |
+
"status": "READY_NOT_STARTED",
|
| 4 |
+
"config_id": "O7B-THINK",
|
| 5 |
+
"arm_id": "O7B-THINK",
|
| 6 |
+
"corrects": "PAIRED_V4",
|
| 7 |
+
"base": {
|
| 8 |
+
"model_id": "allenai/Olmo-3-1025-7B",
|
| 9 |
+
"revision": "a81bae42db3975be1671e27b9c9a56da1a9f980f",
|
| 10 |
+
"local_path": "/workspace/code-sft-infra/models/olmo-3-1025-7b",
|
| 11 |
+
"pure_base": true,
|
| 12 |
+
"warm_start_adapter": null,
|
| 13 |
+
"tied_embeddings": false,
|
| 14 |
+
"tokenizer_sha256": "73fd5254624f39a88e3faac6a8e11300fc3c735ed37880d4f4f08db898eaecca",
|
| 15 |
+
"tokenizer_config_sha256": null
|
| 16 |
+
},
|
| 17 |
+
"boundary": {
|
| 18 |
+
"mode": "trainable_token_rows_both_sides",
|
| 19 |
+
"token_ids": [
|
| 20 |
+
100257
|
| 21 |
+
],
|
| 22 |
+
"token_names": {
|
| 23 |
+
"100257": "<|endoftext|>"
|
| 24 |
+
},
|
| 25 |
+
"row_lr": 0.0001,
|
| 26 |
+
"row_weight_decay": 0.0,
|
| 27 |
+
"fp32_master": true,
|
| 28 |
+
"untie_tied_embeddings": true
|
| 29 |
+
},
|
| 30 |
+
"context": {
|
| 31 |
+
"model_context": 32768,
|
| 32 |
+
"packing": false,
|
| 33 |
+
"truncation": false
|
| 34 |
+
},
|
| 35 |
+
"checkpoint": {
|
| 36 |
+
"milestone_updates": [
|
| 37 |
+
484,
|
| 38 |
+
724,
|
| 39 |
+
968
|
| 40 |
+
],
|
| 41 |
+
"checkpoint_target_tokens": [
|
| 42 |
+
31722729,
|
| 43 |
+
47449622,
|
| 44 |
+
63378772
|
| 45 |
+
],
|
| 46 |
+
"save_only_do_not_pick": true,
|
| 47 |
+
"right_side_3_only": true,
|
| 48 |
+
"primary": "2EP_FINAL_ONLY",
|
| 49 |
+
"note": "right-side: first update fully past 1.0ep, 1.5ep, plus 2ep endpoint"
|
| 50 |
+
},
|
| 51 |
+
"data": {
|
| 52 |
+
"path": "/workspace/code-sft-runs/t30b2507-q4b-tn-v2-lr1e4-e1-20260907/data/THINK_PAIRED_V4_2EP_O7B.parquet",
|
| 53 |
+
"sha256": "17554a69f4b99d6d0b0861bece9e959a664084e00dc38ee76a5ba711e2d8da60",
|
| 54 |
+
"unique_rows": 4715,
|
| 55 |
+
"payload_rows": 9430,
|
| 56 |
+
"physical_rows": 9430,
|
| 57 |
+
"epochs": 2,
|
| 58 |
+
"physical_2ep_rule": "concat(selected_v4_table, selected_v4_table) same order; parquet on disk is one 4715-row copy",
|
| 59 |
+
"planned_active_tokens": 63378772,
|
| 60 |
+
"active_tokens_1ep": 31689386,
|
| 61 |
+
"full_tokens_1ep": 34204230,
|
| 62 |
+
"optimizer_updates": 968,
|
| 63 |
+
"census": "/workspace/code-sft-runs/t30b2507-q4b-tn-v2-lr1e4-e1-20260907/data/TOKEN_CENSUS_V4_O7B_THINK.json",
|
| 64 |
+
"duplicate_rows_allowed": true,
|
| 65 |
+
"order_mutation_allowed": false,
|
| 66 |
+
"token_columns_in_parquet": "legacy tail [151645,198] for Qwen; training uses census/v3 render",
|
| 67 |
+
"source_1ep_sha256": "cf37c85413f69339508004611e196e42abbc2bddbd3462c209a1b97e1d820002"
|
| 68 |
+
},
|
| 69 |
+
"model_runtime": {
|
| 70 |
+
"bf16": true,
|
| 71 |
+
"trainable_dtype": "float32",
|
| 72 |
+
"finetuning_type": "lora",
|
| 73 |
+
"lora_rank": 64,
|
| 74 |
+
"lora_alpha": 128,
|
| 75 |
+
"lora_dropout": 0.0,
|
| 76 |
+
"lora_targets": [
|
| 77 |
+
"q_proj",
|
| 78 |
+
"k_proj",
|
| 79 |
+
"v_proj",
|
| 80 |
+
"o_proj",
|
| 81 |
+
"gate_proj",
|
| 82 |
+
"up_proj",
|
| 83 |
+
"down_proj"
|
| 84 |
+
],
|
| 85 |
+
"modules_to_save": null,
|
| 86 |
+
"gradient_checkpointing": true,
|
| 87 |
+
"trainable_token_indices": {
|
| 88 |
+
"embed_tokens": [
|
| 89 |
+
100257
|
| 90 |
+
],
|
| 91 |
+
"lm_head": [
|
| 92 |
+
100257
|
| 93 |
+
]
|
| 94 |
+
}
|
| 95 |
+
},
|
| 96 |
+
"optimization": {
|
| 97 |
+
"active_tokens_per_update_target": 65536,
|
| 98 |
+
"betas": [
|
| 99 |
+
0.9,
|
| 100 |
+
0.95
|
| 101 |
+
],
|
| 102 |
+
"epsilon": 1e-08,
|
| 103 |
+
"learning_rate": 0.0001,
|
| 104 |
+
"max_grad_norm": 1.0,
|
| 105 |
+
"optimizer": "adamw",
|
| 106 |
+
"weight_decay": 0.1,
|
| 107 |
+
"whole_row_microsteps": true,
|
| 108 |
+
"warmup_fraction": 0.06,
|
| 109 |
+
"warmup_target_tokens": 3802726,
|
| 110 |
+
"scheduler": "cosine_by_assistant_target_token_dose",
|
| 111 |
+
"min_lr_ratio": 0.0,
|
| 112 |
+
"scheduler_horizon_optimizer_updates": 968,
|
| 113 |
+
"target_assistant_tokens": 63378772,
|
| 114 |
+
"worker_lr_scale": "rstar_dual_worker_v4._lr_scale"
|
| 115 |
+
},
|
| 116 |
+
"performance_recipe": {
|
| 117 |
+
"eos_weight": 1.0,
|
| 118 |
+
"prompt_mask": -100
|
| 119 |
+
},
|
| 120 |
+
"recipe_worker_fields": {
|
| 121 |
+
"tokens_per_optimizer_update": 65536,
|
| 122 |
+
"max_sequence_length": 32768,
|
| 123 |
+
"packing": {
|
| 124 |
+
"enabled": false
|
| 125 |
+
},
|
| 126 |
+
"repeat_examples": false,
|
| 127 |
+
"note": "2EP is physical concat; do not recycle via repeat_examples"
|
| 128 |
+
},
|
| 129 |
+
"reproducibility": {
|
| 130 |
+
"seed": 42
|
| 131 |
+
},
|
| 132 |
+
"runner": {
|
| 133 |
+
"path": "/workspace/tools/rstar_dual_worker_v4.py",
|
| 134 |
+
"sha256": "97a298e9ee4f0bb571baa88c7d75285ade2ea2a8a9b5875b35db25b5db9b3842",
|
| 135 |
+
"rewrite_forbidden": true
|
| 136 |
+
},
|
| 137 |
+
"template": {
|
| 138 |
+
"id": "olmo3-think",
|
| 139 |
+
"mode": "think",
|
| 140 |
+
"enable_thinking": true,
|
| 141 |
+
"generation_prefill": false,
|
| 142 |
+
"target_contains_full_think": true,
|
| 143 |
+
"canonical_empty_think_pair_in_prompt": false,
|
| 144 |
+
"empty_think_pair_in_loss": false,
|
| 145 |
+
"target_forbids_think_tags": false,
|
| 146 |
+
"olmo_nothink_empty_pair_equivalent": "Think pair lives in the target (no generation-prompt <think> prefill).",
|
| 147 |
+
"chat_template_file": "templates/olmo3-lcb-noprefill.chat_template.jinja",
|
| 148 |
+
"system_prompt_sha256": "no_explicit_system_use_pinned_olmo_template",
|
| 149 |
+
"chat_template_sha256": "olmo3-lcb-noprefill",
|
| 150 |
+
"target_tail": [
|
| 151 |
+
100257
|
| 152 |
+
],
|
| 153 |
+
"evaluation_stop_ids": [
|
| 154 |
+
100257,
|
| 155 |
+
100265
|
| 156 |
+
],
|
| 157 |
+
"eos_token_ids": [
|
| 158 |
+
100257
|
| 159 |
+
],
|
| 160 |
+
"eos_token": "<|endoftext|>",
|
| 161 |
+
"supervision_eos": 100257,
|
| 162 |
+
"forbidden_label_ids": [
|
| 163 |
+
100265
|
| 164 |
+
]
|
| 165 |
+
},
|
| 166 |
+
"eval": {
|
| 167 |
+
"suite": "DEV256",
|
| 168 |
+
"seed": 3407,
|
| 169 |
+
"enable_thinking": true,
|
| 170 |
+
"canonical_empty_think_pair": false,
|
| 171 |
+
"generated_think_tags": null,
|
| 172 |
+
"temperature": 0.6,
|
| 173 |
+
"top_p": 0.95,
|
| 174 |
+
"top_k": 20,
|
| 175 |
+
"model_context": 32768,
|
| 176 |
+
"max_actual": "ctx-prompt-64",
|
| 177 |
+
"stop_ids": [
|
| 178 |
+
100257,
|
| 179 |
+
100265
|
| 180 |
+
],
|
| 181 |
+
"base_must_reeval_same_mode": true,
|
| 182 |
+
"primary_checkpoint": "2EP_FINAL_ONLY",
|
| 183 |
+
"milestones": "appendix_only",
|
| 184 |
+
"cap_rate": "report_only",
|
| 185 |
+
"do_not_execute": true
|
| 186 |
+
}
|
| 187 |
+
}
|