From deb6e3431c0dca8335d4209b566d69bdd1097abc Mon Sep 17 00:00:00 2001 From: aoshen02 Date: Tue, 9 Jun 2026 13:48:20 +0000 Subject: [PATCH 1/2] fix: add coding_agent_rl dependencies to requirements.txt #148 synced the coding_agent_rl subsystem (slime #1923 et al.) but missed the corresponding requirements: anthropic, e2b, openai, openai-agents. Co-Authored-By: Claude Opus 4.6 (1M context) --- requirements.txt | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/requirements.txt b/requirements.txt index aa8cb6d32..b44dc96f1 100644 --- a/requirements.txt +++ b/requirements.txt @@ -1,12 +1,16 @@ accelerate +anthropic blobfile cloudpickle datasets +e2b httpx[http2] mcp[cli] memray # needed for debugging (but is lightweight), we can put it to dev mode when using pyproject.toml numba omegaconf +openai +openai-agents pillow pylatexenc pyyaml From b332ffc2c80e204ad520f93daa0aa7d1e65294d8 Mon Sep 17 00:00:00 2001 From: aoshen02 Date: Tue, 9 Jun 2026 13:54:20 +0000 Subject: [PATCH 2/2] fix: add trailing newlines to model scripts + missing --mtp-num-layers - Add POSIX-required trailing newline to 22 scripts/models/*.sh files - Restore `--mtp-num-layers 1` to scripts/models/mimo-7B-rl.sh (dropped during sync) Co-Authored-By: Claude Opus 4.6 (1M context) --- scripts/models/deepseek-v3.sh | 2 +- scripts/models/glm4-32B.sh | 2 +- scripts/models/glm4-9B.sh | 2 +- scripts/models/glm4.5-106B-A12B.sh | 2 +- scripts/models/glm4.7-30B-A3B.sh | 2 +- scripts/models/glm5-744B-A40B.sh | 2 +- scripts/models/kimi-k2-thinking.sh | 2 +- scripts/models/kimi-k2.sh | 2 +- scripts/models/llama3.1-8B-Instruct.sh | 2 +- scripts/models/llama3.2-3B-Instruct.sh | 2 +- scripts/models/mimo-7B-rl.sh | 1 + scripts/models/moonlight.sh | 2 +- scripts/models/qwen2.5-0.5B.sh | 2 +- scripts/models/qwen2.5-1.5B.sh | 2 +- scripts/models/qwen2.5-32B.sh | 2 +- scripts/models/qwen2.5-7B.sh | 2 +- scripts/models/qwen3-0.6B.sh | 2 +- scripts/models/qwen3-1.7B.sh | 2 +- scripts/models/qwen3-235B-A22B.sh | 2 +- scripts/models/qwen3-30B-A3B.sh | 2 +- scripts/models/qwen3-4B-Instruct-2507.sh | 2 +- scripts/models/qwen3-4B.sh | 2 +- scripts/models/qwen3-8B.sh | 2 +- 23 files changed, 23 insertions(+), 22 deletions(-) diff --git a/scripts/models/deepseek-v3.sh b/scripts/models/deepseek-v3.sh index 8c50d2c94..13169610f 100644 --- a/scripts/models/deepseek-v3.sh +++ b/scripts/models/deepseek-v3.sh @@ -60,4 +60,4 @@ MODEL_ARGS=( --moe-router-topk-scaling-factor 2.5 --moe-router-dtype fp32 --moe-permute-fusion -) \ No newline at end of file +) diff --git a/scripts/models/glm4-32B.sh b/scripts/models/glm4-32B.sh index ff1d33814..20df4a40f 100644 --- a/scripts/models/glm4-32B.sh +++ b/scripts/models/glm4-32B.sh @@ -21,4 +21,4 @@ MODEL_ARGS=( --rotary-percent 0.5 --no-rope-fusion --untie-embeddings-and-output-weights -) \ No newline at end of file +) diff --git a/scripts/models/glm4-9B.sh b/scripts/models/glm4-9B.sh index e7bab92bc..b3e0dfe05 100644 --- a/scripts/models/glm4-9B.sh +++ b/scripts/models/glm4-9B.sh @@ -20,4 +20,4 @@ MODEL_ARGS=( --rotary-percent 0.5 --no-rope-fusion --untie-embeddings-and-output-weights -) \ No newline at end of file +) diff --git a/scripts/models/glm4.5-106B-A12B.sh b/scripts/models/glm4.5-106B-A12B.sh index 6bab15b77..f5f87362d 100644 --- a/scripts/models/glm4.5-106B-A12B.sh +++ b/scripts/models/glm4.5-106B-A12B.sh @@ -37,4 +37,4 @@ MODEL_ARGS=( --moe-router-dtype fp32 --moe-permute-fusion --moe-aux-loss-coeff 0 -) \ No newline at end of file +) diff --git a/scripts/models/glm4.7-30B-A3B.sh b/scripts/models/glm4.7-30B-A3B.sh index 696eb9bcf..816629721 100644 --- a/scripts/models/glm4.7-30B-A3B.sh +++ b/scripts/models/glm4.7-30B-A3B.sh @@ -49,4 +49,4 @@ MODEL_ARGS=( --vocab-size 154880 --rotary-base 1000000 --no-rope-fusion -) \ No newline at end of file +) diff --git a/scripts/models/glm5-744B-A40B.sh b/scripts/models/glm5-744B-A40B.sh index fd170f000..92c351fc7 100644 --- a/scripts/models/glm5-744B-A40B.sh +++ b/scripts/models/glm5-744B-A40B.sh @@ -52,4 +52,4 @@ MODEL_ARGS=( # vime specific args --allgather-cp -) \ No newline at end of file +) diff --git a/scripts/models/kimi-k2-thinking.sh b/scripts/models/kimi-k2-thinking.sh index b7fceda59..c25d3831d 100644 --- a/scripts/models/kimi-k2-thinking.sh +++ b/scripts/models/kimi-k2-thinking.sh @@ -60,4 +60,4 @@ MODEL_ARGS=( --moe-router-topk-scaling-factor 2.827 --moe-router-dtype fp32 --moe-permute-fusion -) \ No newline at end of file +) diff --git a/scripts/models/kimi-k2.sh b/scripts/models/kimi-k2.sh index eafb7acad..9ef591625 100644 --- a/scripts/models/kimi-k2.sh +++ b/scripts/models/kimi-k2.sh @@ -60,4 +60,4 @@ MODEL_ARGS=( --moe-router-topk-scaling-factor 2.827 --moe-router-dtype fp32 --moe-permute-fusion -) \ No newline at end of file +) diff --git a/scripts/models/llama3.1-8B-Instruct.sh b/scripts/models/llama3.1-8B-Instruct.sh index 0815b3e0a..b092cb83a 100644 --- a/scripts/models/llama3.1-8B-Instruct.sh +++ b/scripts/models/llama3.1-8B-Instruct.sh @@ -17,4 +17,4 @@ MODEL_ARGS=( --use-rope-scaling --rotary-scaling-factor 8.0 --untie-embeddings-and-output-weights -) \ No newline at end of file +) diff --git a/scripts/models/llama3.2-3B-Instruct.sh b/scripts/models/llama3.2-3B-Instruct.sh index ff50130c8..654de5a38 100644 --- a/scripts/models/llama3.2-3B-Instruct.sh +++ b/scripts/models/llama3.2-3B-Instruct.sh @@ -16,4 +16,4 @@ MODEL_ARGS=( --kv-channels 128 --use-rope-scaling --rotary-scaling-factor 32.0 -) \ No newline at end of file +) diff --git a/scripts/models/mimo-7B-rl.sh b/scripts/models/mimo-7B-rl.sh index 3def88224..22366935f 100644 --- a/scripts/models/mimo-7B-rl.sh +++ b/scripts/models/mimo-7B-rl.sh @@ -15,4 +15,5 @@ MODEL_ARGS=( --vocab-size 151680 --untie-embeddings-and-output-weights --max-position-embeddings 32768 + --mtp-num-layers 1 ) diff --git a/scripts/models/moonlight.sh b/scripts/models/moonlight.sh index bcce99892..87633d34f 100644 --- a/scripts/models/moonlight.sh +++ b/scripts/models/moonlight.sh @@ -66,4 +66,4 @@ MODEL_ARGS=( --moe-token-drop-policy probs --moe-router-dtype fp32 --moe-permute-fusion -) \ No newline at end of file +) diff --git a/scripts/models/qwen2.5-0.5B.sh b/scripts/models/qwen2.5-0.5B.sh index 66d5b29a0..6af56e6be 100644 --- a/scripts/models/qwen2.5-0.5B.sh +++ b/scripts/models/qwen2.5-0.5B.sh @@ -13,4 +13,4 @@ MODEL_ARGS=( --group-query-attention --num-query-groups 2 --vocab-size 151936 -) \ No newline at end of file +) diff --git a/scripts/models/qwen2.5-1.5B.sh b/scripts/models/qwen2.5-1.5B.sh index b046a95c6..61d036972 100644 --- a/scripts/models/qwen2.5-1.5B.sh +++ b/scripts/models/qwen2.5-1.5B.sh @@ -13,4 +13,4 @@ MODEL_ARGS=( --group-query-attention --num-query-groups 2 --vocab-size 151936 -) \ No newline at end of file +) diff --git a/scripts/models/qwen2.5-32B.sh b/scripts/models/qwen2.5-32B.sh index 26b49845a..fe4b1519c 100644 --- a/scripts/models/qwen2.5-32B.sh +++ b/scripts/models/qwen2.5-32B.sh @@ -14,4 +14,4 @@ MODEL_ARGS=( --rotary-base 1000000 --vocab-size 152064 --untie-embeddings-and-output-weights -) \ No newline at end of file +) diff --git a/scripts/models/qwen2.5-7B.sh b/scripts/models/qwen2.5-7B.sh index eba912b1d..d43da633c 100644 --- a/scripts/models/qwen2.5-7B.sh +++ b/scripts/models/qwen2.5-7B.sh @@ -14,4 +14,4 @@ MODEL_ARGS=( --rotary-base 1000000 --vocab-size 152064 --untie-embeddings-and-output-weights -) \ No newline at end of file +) diff --git a/scripts/models/qwen3-0.6B.sh b/scripts/models/qwen3-0.6B.sh index f484ec951..e2f813104 100644 --- a/scripts/models/qwen3-0.6B.sh +++ b/scripts/models/qwen3-0.6B.sh @@ -14,4 +14,4 @@ MODEL_ARGS=( --vocab-size 151936 --kv-channels 128 --qk-layernorm -) \ No newline at end of file +) diff --git a/scripts/models/qwen3-1.7B.sh b/scripts/models/qwen3-1.7B.sh index 743599633..3c18442a3 100644 --- a/scripts/models/qwen3-1.7B.sh +++ b/scripts/models/qwen3-1.7B.sh @@ -14,4 +14,4 @@ MODEL_ARGS=( --vocab-size 151936 --kv-channels 128 --qk-layernorm -) \ No newline at end of file +) diff --git a/scripts/models/qwen3-235B-A22B.sh b/scripts/models/qwen3-235B-A22B.sh index 1f6635526..c2a9ccaa2 100644 --- a/scripts/models/qwen3-235B-A22B.sh +++ b/scripts/models/qwen3-235B-A22B.sh @@ -46,4 +46,4 @@ MODEL_ARGS=( --moe-router-dtype fp32 --moe-permute-fusion --moe-aux-loss-coeff 0 -) \ No newline at end of file +) diff --git a/scripts/models/qwen3-30B-A3B.sh b/scripts/models/qwen3-30B-A3B.sh index dfab8682a..37b094329 100644 --- a/scripts/models/qwen3-30B-A3B.sh +++ b/scripts/models/qwen3-30B-A3B.sh @@ -46,4 +46,4 @@ MODEL_ARGS=( --moe-router-dtype fp32 --moe-permute-fusion --moe-aux-loss-coeff 0 -) \ No newline at end of file +) diff --git a/scripts/models/qwen3-4B-Instruct-2507.sh b/scripts/models/qwen3-4B-Instruct-2507.sh index 67d13c0c8..e5de17514 100644 --- a/scripts/models/qwen3-4B-Instruct-2507.sh +++ b/scripts/models/qwen3-4B-Instruct-2507.sh @@ -1 +1 @@ -MODEL_ARGS_ROTARY_BASE=5000000 source "$(dirname -- "${BASH_SOURCE[0]}")/qwen3-4B.sh" \ No newline at end of file +MODEL_ARGS_ROTARY_BASE=5000000 source "$(dirname -- "${BASH_SOURCE[0]}")/qwen3-4B.sh" diff --git a/scripts/models/qwen3-4B.sh b/scripts/models/qwen3-4B.sh index 51f9e4758..67c6235f0 100644 --- a/scripts/models/qwen3-4B.sh +++ b/scripts/models/qwen3-4B.sh @@ -14,4 +14,4 @@ MODEL_ARGS=( --vocab-size 151936 --kv-channels 128 --qk-layernorm -) \ No newline at end of file +) diff --git a/scripts/models/qwen3-8B.sh b/scripts/models/qwen3-8B.sh index fc573adb3..087e10f23 100644 --- a/scripts/models/qwen3-8B.sh +++ b/scripts/models/qwen3-8B.sh @@ -15,4 +15,4 @@ MODEL_ARGS=( --kv-channels 128 --qk-layernorm --untie-embeddings-and-output-weights -) \ No newline at end of file +)