more configs

Author: pawalt

slime/code_golf_harbor_modal/configs/base.py

@@ -70,7 +70,7 @@ def generate_train_args(
 DEFAULT_TRAINING_ARGS = """
     --tensor-model-parallel-size 2 --sequence-parallel
     --recompute-granularity full --recompute-method uniform --recompute-num-layers 1
-    --use-dynamic-batch-size --max-tokens-per-gpu 9216
+    --use-dynamic-batch-size --max-tokens-per-gpu 4096
     --megatron-to-hf-mode bridge
     --attention-dropout 0.0 --hidden-dropout 0.0
     --accumulate-allreduce-grads-in-fp32 --attention-softmax-in-fp32

slime/code_golf_harbor_modal/configs/qwen_8b_multi.py

@@ -34,15 +34,15 @@ def get_config() -> RLConfig:
             --input-key messages
             --label-key label
             --apply-chat-template
-            --apply-chat-template-kwargs '{{"enable_thinking": false}}'
+            --apply-chat-template-kwargs '{{"enable_thinking": true}}'
             --prompt-data {{data_path}}/mbpp_harbor/slime/train.parquet
             --eval-prompt-data mbpp {{data_path}}/mbpp_harbor/slime/test.parquet
 
             # Rollout / batching
             --num-rollout 2000
-            --rollout-batch-size 32
+            --rollout-batch-size 128
             --n-samples-per-prompt 8
-            --global-batch-size 256
+            --global-batch-size 1024
             --rollout-max-response-len 1024
             --rollout-temperature 0.9
             --eval-max-response-len 1024
@@ -54,7 +54,7 @@ def get_config() -> RLConfig:
 
             # SGLang rollout engines
             --rollout-num-gpus-per-engine 2
-            --sglang-mem-fraction-static 0.7
+            --sglang-mem-fraction-static 0.25
 
             # Distributed orchestration
             --actor-num-nodes 4
@@ -66,7 +66,7 @@ def get_config() -> RLConfig:
             --eval-top-p 1
 
             # Save checkpoints to volume
-            --save {{checkpoints_path}}/qwen8b_code_golf
+            --save {{checkpoints_path}}/qwen8b_code_golf_thinking
             --save-interval 20
         """,
     )

slime/code_golf_harbor_modal/configs/qwen_8b_multi_4k.py

@@ -0,0 +1,72 @@
+from __future__ import annotations
+
+from .base import (
+    DEFAULT_GRPO_ARGS,
+    DEFAULT_OPTIMIZER_ARGS,
+    DEFAULT_TRAINING_ARGS,
+    QWEN3_8B_MODEL_ARGS,
+    RLConfig,
+)
+
+
+def get_config() -> RLConfig:
+    return RLConfig(
+        model_name="Qwen3-8B",
+        model_id="Qwen/Qwen3-8B",
+        app_name="slime-qwen8b-code-golf",
+        n_nodes=4,
+        gpu="H100:8",
+        sync=True,
+        wandb_project="slime-code-golf",
+        wandb_run_name_prefix="qwen8b-mbpp-harbor-thinking-4k",
+        harbor_rm_profile=True,
+        harbor_rm_log_samples=True,
+        slime_args=f"""
+            # Model
+            {QWEN3_8B_MODEL_ARGS}
+
+            # Training + optimizer + GRPO
+            {DEFAULT_TRAINING_ARGS}
+            {DEFAULT_OPTIMIZER_ARGS}
+            {DEFAULT_GRPO_ARGS}
+
+            # Dataset format
+            --input-key messages
+            --label-key label
+            --apply-chat-template
+            --apply-chat-template-kwargs '{{"enable_thinking": true}}'
+            --prompt-data {{data_path}}/mbpp_harbor/slime/train.parquet
+            --eval-prompt-data mbpp {{data_path}}/mbpp_harbor/slime/test.parquet
+
+            # Rollout / batching
+            --num-rollout 2000
+            --rollout-batch-size 128
+            --n-samples-per-prompt 8
+            --global-batch-size 1024
+            --rollout-max-response-len 2048
+            --rollout-temperature 0.9
+            --eval-max-response-len 2048
+            --n-samples-per-eval-prompt 8
+
+            # Custom reward model (Harbor + Modal sandbox scoring)
+            --rm-type math
+            --custom-rm-path custom_rm.custom_rm
+
+            # SGLang rollout engines
+            --rollout-num-gpus-per-engine 2
+            --sglang-mem-fraction-static 0.7
+
+            # Distributed orchestration
+            --actor-num-nodes 4
+            --actor-num-gpus-per-node 8
+            --colocate
+
+            # Eval cadence
+            --eval-interval 20
+            --eval-top-p 1
+
+            # Save checkpoints to volume
+            --save {{checkpoints_path}}/qwen8b_code_golf_thinking_4k
+            --save-interval 20
+        """,
+    )

slime/code_golf_harbor_modal/configs/qwen_8b_multi_4k_1024.py

@@ -0,0 +1,72 @@
+from __future__ import annotations
+
+from .base import (
+    DEFAULT_GRPO_ARGS,
+    DEFAULT_OPTIMIZER_ARGS,
+    DEFAULT_TRAINING_ARGS,
+    QWEN3_8B_MODEL_ARGS,
+    RLConfig,
+)
+
+
+def get_config() -> RLConfig:
+    return RLConfig(
+        model_name="Qwen3-8B",
+        model_id="Qwen/Qwen3-8B",
+        app_name="slime-qwen8b-code-golf",
+        n_nodes=4,
+        gpu="H100:8",
+        sync=True,
+        wandb_project="slime-code-golf",
+        wandb_run_name_prefix="qwen8b-mbpp-harbor-thinking-4k-1024",
+        harbor_rm_profile=True,
+        harbor_rm_log_samples=True,
+        slime_args=f"""
+            # Model
+            {QWEN3_8B_MODEL_ARGS}
+
+            # Training + optimizer + GRPO
+            {DEFAULT_TRAINING_ARGS}
+            {DEFAULT_OPTIMIZER_ARGS}
+            {DEFAULT_GRPO_ARGS}
+
+            # Dataset format
+            --input-key messages
+            --label-key label
+            --apply-chat-template
+            --apply-chat-template-kwargs '{{"enable_thinking": true}}'
+            --prompt-data {{data_path}}/mbpp_harbor/slime/train.parquet
+            --eval-prompt-data mbpp {{data_path}}/mbpp_harbor/slime/test.parquet
+
+            # Rollout / batching (128 * 8 = 1024 samples)
+            --num-rollout 2000
+            --rollout-batch-size 128
+            --n-samples-per-prompt 8
+            --global-batch-size 1024
+            --rollout-max-response-len 4096
+            --rollout-temperature 0.9
+            --eval-max-response-len 4096
+            --n-samples-per-eval-prompt 8
+
+            # Custom reward model (Harbor + Modal sandbox scoring)
+            --rm-type math
+            --custom-rm-path custom_rm.custom_rm
+
+            # SGLang rollout engines
+            --rollout-num-gpus-per-engine 2
+            --sglang-mem-fraction-static 0.7
+
+            # Distributed orchestration
+            --actor-num-nodes 4
+            --actor-num-gpus-per-node 8
+            --colocate
+
+            # Eval cadence
+            --eval-interval 20
+            --eval-top-p 1
+
+            # Save checkpoints to volume
+            --save {{checkpoints_path}}/qwen8b_code_golf_thinking_4k_1024
+            --save-interval 20
+        """,
+    )

slime/code_golf_harbor_modal/configs/qwen_8b_multi_4k_512.py

@@ -0,0 +1,72 @@
+from __future__ import annotations
+
+from .base import (
+    DEFAULT_GRPO_ARGS,
+    DEFAULT_OPTIMIZER_ARGS,
+    DEFAULT_TRAINING_ARGS,
+    QWEN3_8B_MODEL_ARGS,
+    RLConfig,
+)
+
+
+def get_config() -> RLConfig:
+    return RLConfig(
+        model_name="Qwen3-8B",
+        model_id="Qwen/Qwen3-8B",
+        app_name="slime-qwen8b-code-golf",
+        n_nodes=4,
+        gpu="H100:8",
+        sync=True,
+        wandb_project="slime-code-golf",
+        wandb_run_name_prefix="qwen8b-mbpp-harbor-thinking-4k-512",
+        harbor_rm_profile=True,
+        harbor_rm_log_samples=True,
+        slime_args=f"""
+            # Model
+            {QWEN3_8B_MODEL_ARGS}
+
+            # Training + optimizer + GRPO
+            {DEFAULT_TRAINING_ARGS}
+            {DEFAULT_OPTIMIZER_ARGS}
+            {DEFAULT_GRPO_ARGS}
+
+            # Dataset format
+            --input-key messages
+            --label-key label
+            --apply-chat-template
+            --apply-chat-template-kwargs '{{"enable_thinking": true}}'
+            --prompt-data {{data_path}}/mbpp_harbor/slime/train.parquet
+            --eval-prompt-data mbpp {{data_path}}/mbpp_harbor/slime/test.parquet
+
+            # Rollout / batching (64 * 8 = 512 samples)
+            --num-rollout 2000
+            --rollout-batch-size 64
+            --n-samples-per-prompt 8
+            --global-batch-size 512
+            --rollout-max-response-len 4096
+            --rollout-temperature 0.9
+            --eval-max-response-len 4096
+            --n-samples-per-eval-prompt 8
+
+            # Custom reward model (Harbor + Modal sandbox scoring)
+            --rm-type math
+            --custom-rm-path custom_rm.custom_rm
+
+            # SGLang rollout engines
+            --rollout-num-gpus-per-engine 2
+            --sglang-mem-fraction-static 0.7
+
+            # Distributed orchestration
+            --actor-num-nodes 4
+            --actor-num-gpus-per-node 8
+            --colocate
+
+            # Eval cadence
+            --eval-interval 20
+            --eval-top-p 1
+
+            # Save checkpoints to volume
+            --save {{checkpoints_path}}/qwen8b_code_golf_thinking_4k_512
+            --save-interval 20
+        """,
+    )