Downloads · 30 days
4
11% of all-time downloads
yujackein/onereason-8b-lora-item32k-user100-rec50-worldclean1601-all1-lr2e4-r32a32-seq32768-gbs4-1epoch
onereason-8b-lora-item32k-user100-rec50-worldclean1601-all1-lr2e4-r32a32-seq32768-gbs4-1epoch is a text generation model from yujackein. Use it when you need the model to write or continue text. It is set up for peft.
This is an experimental LoRA adapter for OpenOneRec/OneReason-8B-pretrain-competition, trained for the OneReason recommendation competition.
Downloads · 30 days
4
11% of all-time downloads
All-time downloads
36
Public
Repo size
365 MB
Likes
0
Public
Click a slice to open those files.
.safetensors349 MB · 96%
From the Hugging Face model README
This is an experimental LoRA adapter for
OpenOneRec/OneReason-8B-pretrain-competition,
trained for the OneReason recommendation competition.
This is the completed GBS4 run. It must not be confused with the aborted GBS16 contract check, which stopped at step 13 and produced no checkpoint.
This card extends the Item32K / User75 / Rec50 / WorldClean1601 recipe by using all official R2 evolution rows. The previous User75 selection is a strict subset; R0, R3 and world rows are unchanged.
| Category | Rows | Rendered tokens |
|---|---|---|
| R0 perception | 32,000 | 5,801,468 |
| R2 evolution | 32,848 | 158,851,833 |
| R3 recommendation | 24,134 | 34,445,219 |
| Clean C-Eval world data | 1,601 | 195,253 |
| Total | 90,583 | 199,293,773 |
Relative to User75, this adds 8,212 official R2 rows and 39,720,865 rendered tokens. All assistant loss weights are 1.0. No row exceeds the 32,768-token cutoff. The four-rank packing plan consumes 99.2838% of rendered tokens in one safe pass.
r=32, alpha=32, dropout 0.05)q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj2e-4, cosine schedule, 3% warmupAdapter SHA-256:
ed29cafae7cabb8fa364f6b989c7a8a6c7a2c88d0c019b466947d251d377ce70
import torch
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
base_id = "OpenOneRec/OneReason-8B-pretrain-competition"
adapter_id = "yujackein/onereason-8b-lora-item32k-user100-rec50-worldclean1601-all1-lr2e4-r32a32-seq32768-gbs4-1epoch"
tokenizer = AutoTokenizer.from_pretrained(adapter_id, trust_remote_code=True)
base_model = AutoModelForCausalLM.from_pretrained(
base_id,
torch_dtype=torch.bfloat16,
device_map="auto",
trust_remote_code=True,
)
model = PeftModel.from_pretrained(base_model, adapter_id)
Formal platform evaluation results have not yet been added. The 1.2303
score belongs to the User75 / Rec50 / WorldClean1601 LR2e-4 reference model,
not to this User100 adapter.