biondizzle

biondizzle created repository biondizzle/vllm-to-sglang

2026-04-11 23:37:08 +00:00

biondizzle pushed to cuda-malloc-managed at biondizzle/grace-gpu-containers

2026-04-11 23:19:50 +00:00

be4198e754 Add CMM_BUILD_DATE cache-bust arg to Dockerfile

biondizzle pushed to cuda-malloc-managed at biondizzle/grace-gpu-containers

2026-04-11 02:15:10 +00:00

bcc872c2c3 Remove global allocator swap, use targeted KV cache managed allocation

biondizzle pushed to cmm at biondizzle/vllm

2026-04-11 02:14:36 +00:00

7f35bc4158 Targeted KV cache managed memory allocation

biondizzle pushed to cuda-malloc-managed at biondizzle/grace-gpu-containers

2026-04-10 18:37:12 +00:00

07468031db Sync managed_alloc.cu: selective prefetch (<2 GiB to GPU)

biondizzle pushed to master at biondizzle/smollora

2026-04-10 17:35:10 +00:00

278d87286a Remove broken in-place replacement code, use _tokenize_with_special_tokens only

biondizzle pushed to master at biondizzle/smollora

2026-04-10 17:28:53 +00:00

ca50973065 Fix tokenization: replace text token sequences with actual special token IDs

biondizzle pushed to master at biondizzle/smollora

2026-04-10 17:20:07 +00:00

d3b5f04f88 Add embed_tokens to LoRA targets + token ID verification before training

biondizzle pushed to master at biondizzle/smollora

2026-04-10 17:14:58 +00:00

f46995690c Update runbook for tool-call token training run

biondizzle pushed to master at biondizzle/smollora

2026-04-10 17:07:31 +00:00

af497eb16c Add training plan: teach model to emit native tool-call tokens

biondizzle pushed to master at biondizzle/smollora

2026-04-10 16:52:09 +00:00

d1e8c306e3 Add critical training objective: teach model to emit native tool-call tokens

biondizzle pushed to master at biondizzle/smollm3-3b-vllm

2026-04-10 16:51:44 +00:00

bcdbe08037 Document model's inability to emit native tool-call tokens

biondizzle pushed to master at biondizzle/chat-template-debugger

2026-04-10 16:17:22 +00:00

6e44836af6 Support prompt file as positional arg in stage1

biondizzle pushed to master at biondizzle/chat-template-debugger

2026-04-10 16:11:57 +00:00

becee624c6 Add stage1b - API debugger with chat template

biondizzle pushed to master at biondizzle/chat-template-debugger

2026-04-10 15:36:47 +00:00

f20d3bebc3 Default to write_file prompt for stage1

biondizzle pushed to master at biondizzle/chat-template-debugger

2026-04-10 15:33:10 +00:00

9a6fbd5797 Add write_file prompt from model-tool-tests

biondizzle pushed to cuda-malloc-managed at biondizzle/grace-gpu-containers

2026-04-10 15:32:17 +00:00

cdfd37c1e6 Fix Dockerfile: separate git clone and build RUN commands

biondizzle created branch master in biondizzle/chat-template-debugger

2026-04-10 15:28:44 +00:00

biondizzle pushed to master at biondizzle/chat-template-debugger

2026-04-10 15:28:44 +00:00

c981416dde Initial chat template debugger - vLLM raw token inspector

biondizzle created repository biondizzle/chat-template-debugger

2026-04-10 15:06:32 +00:00