Independent publisher
Fahim Tajwar
ftajwar
LLMs, RLHF
Compare the published evidence
This profile connects the 12 model and dataset records shown below to their upstream cards and sources. The library counts describe SAVRN's coverage, rather than this publisher's entire output.
Compare model specifications, source dates, and hardware estimates on the individual records. Publisher identity verification is separate from a model's quality, deployment readiness, or suitability for your work.
Models
Datasets
Training rollouts of one verl run, one parquet per trainer step (data/step-.parquet), rows in the order the trainer consumed them; the rollout.n samples of a prompt are contiguous and share uid. prompttokenids / responsetokenids are the exact token ids (special tokens kept, no padding) and rolloutlogprobs the sampling-time log-prob of every response token. input / output are decoded with skipspecialtokens=True and are for reading only. score is the raw reward.
Training rollouts of one verl run, one parquet per trainer step (data/step-.parquet), rows in the order the trainer consumed them; the rollout.n samples of a prompt are contiguous and share uid. prompttokenids / responsetokenids are the exact token ids (special tokens kept, no padding) and rolloutlogprobs the sampling-time log-prob of every response token. input / output are decoded with skipspecialtokens=True and are for reading only. score is the raw reward.
Training rollouts of one verl run, one parquet per trainer step (data/step-.parquet), rows in the order the trainer consumed them; the rollout.n samples of a prompt are contiguous and share uid. prompttokenids / responsetokenids are the exact token ids (special tokens kept, no padding) and rolloutlogprobs the sampling-time log-prob of every response token. input / output are decoded with skipspecialtokens=True and are for reading only. score is the raw reward.
Training rollouts of one verl run, one parquet per trainer step (data/step-.parquet), rows in the order the trainer consumed them; the rollout.n samples of a prompt are contiguous and share uid. prompttokenids / responsetokenids are the exact token ids (special tokens kept, no padding) and rolloutlogprobs the sampling-time log-prob of every response token. input / output are decoded with skipspecialtokens=True and are for reading only. score is the raw reward.
Training rollouts of one verl run, one parquet per trainer step (data/step-.parquet), rows in the order the trainer consumed them; the rollout.n samples of a prompt are contiguous and share uid. prompttokenids / responsetokenids are the exact token ids (special tokens kept, no padding) and rolloutlogprobs the sampling-time log-prob of every response token. input / output are decoded with skipspecialtokens=True and are for reading only. score is the raw reward.
Training rollouts of one verl run, one parquet per trainer step (data/step-.parquet), rows in the order the trainer consumed them; the rollout.n samples of a prompt are contiguous and share uid. prompttokenids / responsetokenids are the exact token ids (special tokens kept, no padding) and rolloutlogprobs the sampling-time log-prob of every response token. input / output are decoded with skipspecialtokens=True and are for reading only. score is the raw reward.
Training rollouts of one verl run, one parquet per trainer step (data/step-.parquet), rows in the order the trainer consumed them; the rollout.n samples of a prompt are contiguous and share uid. prompttokenids / responsetokenids are the exact token ids (special tokens kept, no padding) and rolloutlogprobs the sampling-time log-prob of every response token. input / output are decoded with skipspecialtokens=True and are for reading only. score is the raw reward.
Training rollouts of one verl run, one parquet per trainer step (data/step-.parquet), rows in the order the trainer consumed them; the rollout.n samples of a prompt are contiguous and share uid. prompttokenids / responsetokenids are the exact token ids (special tokens kept, no padding) and rolloutlogprobs the sampling-time log-prob of every response token. input / output are decoded with skipspecialtokens=True and are for reading only. score is the raw reward.