Super tiny add Qwen3-4B-Instruct-2507 model config (#661)

fzyzcjy · web-flow · commit 9aa8d246c072 · 2025-11-14T16:03:24.000+08:00
diff --git a/scripts/models/qwen3-4B-Instruct-2507.sh b/scripts/models/qwen3-4B-Instruct-2507.sh
@@ -0,0 +1 @@
+MODEL_ARGS_ROTARY_BASE=5000000 source "$(dirname "$0")/qwen3-4B.sh"
diff --git a/scripts/models/qwen3-4B.sh b/scripts/models/qwen3-4B.sh
@@ -10,7 +10,7 @@ MODEL_ARGS=(
    --disable-bias-linear
    --normalization "RMSNorm"
    --norm-epsilon 1e-6
-   --rotary-base 1000000
+   --rotary-base "${MODEL_ARGS_ROTARY_BASE:-1000000}"
    --vocab-size 151936
    --kv-channels 128
    --qk-layernorm

Original file line number	Diff line number	Diff line change
`@@ -0,0 +1 @@`
	`1`	`+MODEL_ARGS_ROTARY_BASE=5000000 source "$(dirname "$0")/qwen3-4B.sh"`