Text Generation
Transformers
Safetensors
English
testgeniy
causal-lm
reasoning
mathematics
logic
long-context
4k-context
small-language-model
Instructions to use Asilarkness/testgeniy with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use Asilarkness/testgeniy with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="Asilarkness/testgeniy")# Load model directly from transformers import AutoModelForCausalLM model = AutoModelForCausalLM.from_pretrained("Asilarkness/testgeniy", device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use Asilarkness/testgeniy with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "Asilarkness/testgeniy" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Asilarkness/testgeniy", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/Asilarkness/testgeniy
- SGLang
How to use Asilarkness/testgeniy 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 "Asilarkness/testgeniy" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Asilarkness/testgeniy", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'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 "Asilarkness/testgeniy" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Asilarkness/testgeniy", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use Asilarkness/testgeniy with Docker Model Runner:
docker model run hf.co/Asilarkness/testgeniy
Add v3 evaluation artifact V3_FINAL_DECISION.json
Browse files
candidates/budgie-alignment-v2/reasoning-sft-then-rl-v3/recovery/V3_FINAL_DECISION.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"date": "2026-08-26",
|
| 3 |
+
"stage": "reasoning SFT then RL v3",
|
| 4 |
+
"status": "Stage A rejected; Stage B RL not started; leader unchanged",
|
| 5 |
+
"base": "verified-math-a025",
|
| 6 |
+
"data": {
|
| 7 |
+
"scanned": 5251,
|
| 8 |
+
"verified_pool": 1600,
|
| 9 |
+
"train": 800,
|
| 10 |
+
"dev": 208,
|
| 11 |
+
"difficulty_train": {
|
| 12 |
+
"medium": 511,
|
| 13 |
+
"hard": 289
|
| 14 |
+
},
|
| 15 |
+
"sources_train": {
|
| 16 |
+
"olympiads": 357,
|
| 17 |
+
"aops_forum": 225,
|
| 18 |
+
"cn_contest": 154,
|
| 19 |
+
"amc_aime": 32,
|
| 20 |
+
"inequalities": 19,
|
| 21 |
+
"olympiads_ref": 10,
|
| 22 |
+
"number_theory": 3
|
| 23 |
+
},
|
| 24 |
+
"problem_types_train": {
|
| 25 |
+
"Number Theory": 179,
|
| 26 |
+
"Algebra": 228,
|
| 27 |
+
"Geometry": 193,
|
| 28 |
+
"Combinatorics": 119,
|
| 29 |
+
"Calculus": 19,
|
| 30 |
+
"Logic and Puzzles": 29,
|
| 31 |
+
"Other": 11,
|
| 32 |
+
"Inequalities": 22
|
| 33 |
+
},
|
| 34 |
+
"rejections": {
|
| 35 |
+
"verify": 2910,
|
| 36 |
+
"length": 740
|
| 37 |
+
},
|
| 38 |
+
"benchmark_family_rows": 0,
|
| 39 |
+
"decontamination": "normalized exact + shared fixed-gate 12-token windows + source bans",
|
| 40 |
+
"verification": "math_verify(reference solution, answer)",
|
| 41 |
+
"seconds": 69.16955494880676,
|
| 42 |
+
"difficulty_dev": {
|
| 43 |
+
"hard": 78,
|
| 44 |
+
"medium": 130
|
| 45 |
+
}
|
| 46 |
+
},
|
| 47 |
+
"encoding": {
|
| 48 |
+
"raw_records": 926,
|
| 49 |
+
"encoded": 908,
|
| 50 |
+
"math": 800,
|
| 51 |
+
"info": 108,
|
| 52 |
+
"mean_tokens": 576.6332599118942,
|
| 53 |
+
"mean_response_tokens": 441.8458149779736,
|
| 54 |
+
"max_tokens": 1536,
|
| 55 |
+
"seconds": 14.342991828918457
|
| 56 |
+
},
|
| 57 |
+
"training": {
|
| 58 |
+
"method": "one-epoch long verified reasoning SFT with sampled-action leader KL",
|
| 59 |
+
"base": "verified-math-a025",
|
| 60 |
+
"encoded_rows": 908,
|
| 61 |
+
"math_rows": 800,
|
| 62 |
+
"info_rows": 108,
|
| 63 |
+
"updates": 227,
|
| 64 |
+
"grad_accum": 4,
|
| 65 |
+
"peak_lr": 3e-07,
|
| 66 |
+
"kl_beta": 0.12,
|
| 67 |
+
"max_length": 1536,
|
| 68 |
+
"source_path": "/marimo/budgie_information_rl_v1/v3_reasoning_sft_source.pt",
|
| 69 |
+
"parameter_count_change": 0,
|
| 70 |
+
"benchmark_family_rows": 0
|
| 71 |
+
},
|
| 72 |
+
"results": {
|
| 73 |
+
"information_dev": {
|
| 74 |
+
"leader": "3/21, loops 0",
|
| 75 |
+
"alpha_0_005": "4/21, loops 0",
|
| 76 |
+
"smaller_alphas": "3/21"
|
| 77 |
+
},
|
| 78 |
+
"random_verified_math_dev_32": {
|
| 79 |
+
"leader": "1/32",
|
| 80 |
+
"alpha_0_005": "1/32"
|
| 81 |
+
},
|
| 82 |
+
"fixed_reasoning": {
|
| 83 |
+
"leader": {
|
| 84 |
+
"gsm": "5/30",
|
| 85 |
+
"math": "3/15",
|
| 86 |
+
"arc": "11/30",
|
| 87 |
+
"folio": "13/30"
|
| 88 |
+
},
|
| 89 |
+
"alpha_0_005": {
|
| 90 |
+
"gsm": "4/30",
|
| 91 |
+
"math": "2/15",
|
| 92 |
+
"arc": "11/30",
|
| 93 |
+
"folio": "13/30"
|
| 94 |
+
}
|
| 95 |
+
}
|
| 96 |
+
},
|
| 97 |
+
"decision": "Reject reasoning SFT Stage A because the only alpha with an information-management gain regresses fixed GSM and MATH. Do not run follow-on RL. Smaller alphas have no held-out information gain. Keep verified-math-a025.",
|
| 98 |
+
"parameter_count_change": 0,
|
| 99 |
+
"benchmark_family_rows": 0,
|
| 100 |
+
"public_root_changed": false
|
| 101 |
+
}
|