Downloads · 30 days
0
AxisMeru/prabhasa-nyaya-step2-sft-arms-archive
prabhasa-nyaya-step2-sft-arms-archive is a machine learning model from AxisMeru. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as other.
Superseded step2-sft sweep arms from the prabhasa-nyaya project, base model Qwen/Qwen2.5-3B (Qwen Research licence, non-commercial). These are kept here as a public archive of intermediate sweep results; the project's…
Downloads · 30 days
0
Access
Public
Updated Sep 23, 2026
Repo size
24.7 GB
Likes
0
Public
Click a slice to open those files.
.safetensors24.7 GB · 100%
From the Hugging Face model README
Superseded step2-sft sweep arms from the prabhasa-nyaya project, base model
Qwen/Qwen2.5-3B (Qwen Research licence, non-commercial). These are kept here
as a public archive of intermediate sweep results; the project's own
gates/HANDOFF record which arm(s), if any, fed forward into later decisions.
Each subfolder is one arm's final checkpoint, weights only (model.safetensors,
config.json, tokenizer files) -- no optimizer state, RNG state, or scheduler
state is included, since these arms are not intended to be resumed for further
training.
arm_c_orderrev/ -- t_tier phase, epoch 4, global_step 1947 (final slot per
the run's own pointer.json).arm_c_ordervol_discrim/ -- t_tier phase, epoch 4, global_step 947 (final
slot per the run's own pointer.json).arm_c_v4_rebalanced/ -- p_tier phase, epoch 1, global_step 1455
(model_final/ in the source run).arm_c_v5b/ -- p_tier phase, epoch 1, global_step 1947, record_index 6411
(model_final/ in the source run).Results for these sweep arms are recorded in the project's own gate JSON files and HANDOFF documents, not restated here.