Downloads · 30 days
0
JacopoCirica/gptoss120b-sdpo-exp10c-argmax-revkl-floor10-step120
gptoss120b-sdpo-exp10c-argmax-revkl-floor10-step120 is a machine learning model from JacopoCirica. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Offline self-distillation (argmax:reward + reverse-KL PPO-clip + teacherlpfloor=-10) of openai/gpt-oss-120b on the Mercor APEX-Agents MCP-480 dataset.
Downloads · 30 days
0
Access
Public
Updated Jun 3, 2026
Repo size
10.5 GB
Likes
0
Public
Click a slice to open those files.
.safetensors5.3 GB · 100%
From the Hugging Face model README
Offline self-distillation (argmax:reward + reverse-KL PPO-clip + teacher_lp_floor=-10)
of openai/gpt-oss-120b on the Mercor APEX-Agents MCP-480 dataset.
target_modules="all-linear", dropout=0openai/gpt-oss-120bThese are raw Tinker adapter weights — adapter_model.safetensors uses Tinker's
internal parameter naming, NOT yet remapped to HF/PEFT key names. To serve under vLLM:
from tinker_cookbook import weights
weights.build_lora_adapter(
base_model="openai/gpt-oss-120b",
adapter_path="<this repo>",
output_path="<peft_out>", # PEFT-format, vLLM --lora-modules ready
)
(or build_hf_model(...) to produce a fully merged model)The conversion only reads the base model's safetensors headers (weight-key names) for remapping; it does not retrain anything.