-
200k ctx (rope_scaling hook) + openai-compatible api (key 1234) + bootstrap v2 (chat 7861 / api 7862) + model card
-
200k ctx (rope_scaling hook) + openai-compatible api (key 1234) + bootstrap v2 (chat 7861 / api 7862) + model card
-
200k ctx (rope_scaling hook) + openai-compatible api (key 1234) + bootstrap v2 (chat 7861 / api 7862) + model card
-
200k ctx (rope_scaling hook) + openai-compatible api (key 1234) + bootstrap v2 (chat 7861 / api 7862) + model card
-
200k ctx (rope_scaling hook) + openai-compatible api (key 1234) + bootstrap v2 (chat 7861 / api 7862) + model card
-
one-cell colab bootstrap (deps+model+server+tunnel)
-
post-SFT eval report (math 8/9, original rent 3/3)
-
chat server (flask, port 7861, force_think+antiloop+votes)
-
samai-4b SFT merged full weights (rent-math sft, 831 rows x2ep)
-
sft adapter (lora r16 + router/ponder_head, loss 0.611)
-
samai-4b: spark-x2.5-4b backbone + pondernet/dmoe conversion
-
initial commit