[1/N] Tiny execute Ruff auto lint (#991)

fzyzcjy · web-flow · commit 4582feb5a35e · 2025-12-01T20:16:53.000+08:00
diff --git a/examples/formal_math/single_round/run.py b/examples/formal_math/single_round/run.py
@@ -134,7 +134,7 @@ def execute():
 
     sglang_args = (
         # "--rollout-num-gpus-per-engine 2 "
-        f"--rollout-num-gpus-per-engine 8 "  # temp use 1 engine per node to avoid flashinfer err
+        "--rollout-num-gpus-per-engine 8 "  # temp use 1 engine per node to avoid flashinfer err
         "--sglang-mem-fraction-static 0.7 "
     )
 
diff --git a/examples/search-r1/generate_with_search.py b/examples/search-r1/generate_with_search.py
@@ -134,7 +134,7 @@ async def execute_predictions(prediction: str) -> str:
         next_obs = ""
         done = True
     else:
-        next_obs = f"\nMy previous action is invalid. \
+        next_obs = "\nMy previous action is invalid. \
 If I want to search, I should put the query between <search> and </search>. \
 If I want to give the final answer, I should put the answer between <answer> and </answer>. Let me try again.\n"
         done = False
@@ -143,7 +143,7 @@ async def execute_predictions(prediction: str) -> str:
 
 
 async def generate(args, sample: Sample, sampling_params) -> Sample:
-    assert not args.partial_rollout, f"Partial rollout is not supported for this function at the moment."
+    assert not args.partial_rollout, "Partial rollout is not supported for this function at the moment."
 
     state = GenerateState(args)
 
diff --git a/examples/search-r1/qa_em_format.py b/examples/search-r1/qa_em_format.py
@@ -180,7 +180,7 @@ def compute_score_em(
     do_print = random.randint(1, 64) == 1
 
     if do_print:
-        print(f"--------------------------------")
+        print("--------------------------------")
         print(f"Golden answers: {ground_truth['target']}")
         print(f"Extracted answer: {answer}")
         print(f"Solution string: {solution_str}")
diff --git a/scripts/run_deepseek.py b/scripts/run_deepseek.py
@@ -158,7 +158,7 @@ def train(args: ScriptArgs):
                 "--prompt-data /root/datasets/gsm8k/train.parquet "
                 "--input-key messages "
                 # Deliberately make it very short for this easy task
-                f"--rollout-max-response-len 256 "
+                "--rollout-max-response-len 256 "
             )
             eval_args += (
                 "--eval-prompt-data gsm8k /root/datasets/gsm8k/test.parquet "
diff --git a/scripts/run_glm45_355b_a32b.py b/scripts/run_glm45_355b_a32b.py
@@ -166,7 +166,7 @@ def train(args: ScriptArgs):
                 "--prompt-data /root/datasets/gsm8k/train.parquet "
                 "--input-key messages "
                 # Deliberately make it very short for this easy task
-                f"--rollout-max-response-len 256 "
+                "--rollout-max-response-len 256 "
             )
             eval_args += (
                 "--eval-prompt-data gsm8k /root/datasets/gsm8k/test.parquet "
@@ -179,7 +179,7 @@ def train(args: ScriptArgs):
         perf_args = (
             "--tensor-model-parallel-size 4 "
             "--sequence-parallel "
-            f"--pipeline-model-parallel-size 1 "
+            "--pipeline-model-parallel-size 1 "
             "--context-parallel-size 1 "
             "--expert-model-parallel-size 4 "
             "--expert-tensor-parallel-size 1 "
diff --git a/scripts/run_qwen3_4b.py b/scripts/run_qwen3_4b.py
@@ -160,7 +160,7 @@ def execute(args: ScriptArgs):
         "--adam-beta2 0.98 "
     )
 
-    sglang_args = f"--rollout-num-gpus-per-engine 1 " "--sglang-chunked-prefill-size 4096 "
+    sglang_args = "--rollout-num-gpus-per-engine 1 " "--sglang-chunked-prefill-size 4096 "
 
     match args.train_backend:
         case "fsdp":
@@ -171,7 +171,7 @@ def execute(args: ScriptArgs):
                 f"--update-weight-buffer-size {512 * 1024 * 1024} "  # 512MB
                 """--train-env-vars '{"PYTORCH_CUDA_ALLOC_CONF":"expandable_segments:True"}' """
             )
-            sglang_args += f"--sglang-mem-fraction-static 0.75 "
+            sglang_args += "--sglang-mem-fraction-static 0.75 "
             perf_args = "--use-dynamic-batch-size " "--max-tokens-per-gpu 32768 "
 
         case "megatron":
@@ -196,7 +196,7 @@ def execute(args: ScriptArgs):
                 "--train-memory-margin-bytes 3221225472 "
             )
             # TODO improve
-            sglang_args += f"--sglang-mem-fraction-static 0.7 "
+            sglang_args += "--sglang-mem-fraction-static 0.7 "
             perf_args = "--use-dynamic-batch-size " "--max-tokens-per-gpu 9216 "
 
         case _:
diff --git a/slime/backends/fsdp_utils/models/qwen3_moe.py b/slime/backends/fsdp_utils/models/qwen3_moe.py
@@ -44,7 +44,7 @@ def fused_experts_impl(
 ):
     padded_size = 0
 
-    assert hidden_states.shape[1] == w1.shape[2] - padded_size, f"Hidden size mismatch"
+    assert hidden_states.shape[1] == w1.shape[2] - padded_size, "Hidden size mismatch"
     assert topk_weights.shape == topk_ids.shape, "topk shape mismatch"
     assert hidden_states.is_contiguous(), "Hidden_states must be contiguous"
     assert w1.is_contiguous(), "Expert weights1 must be contiguous"
diff --git a/slime/backends/megatron_utils/arguments.py b/slime/backends/megatron_utils/arguments.py
@@ -24,7 +24,7 @@ def set_default_megatron_args(args):
         args.padded_vocab_size = _vocab_size_with_padding(args.vocab_size, args)
 
     if not args.tokenizer_model and not args.tokenizer_type:
-        logger.info(f"--tokenizer-model not set, use --hf-checkpoint as tokenizer model.")
+        logger.info("--tokenizer-model not set, use --hf-checkpoint as tokenizer model.")
         args.tokenizer_model = args.hf_checkpoint
         args.tokenizer_type = "HuggingFaceTokenizer"
     return args
diff --git a/slime/ray/rollout.py b/slime/ray/rollout.py
@@ -522,7 +522,7 @@ def _log_rollout_data(rollout_id, args, samples, rollout_extra_metrics, rollout_
     if args.rollout_num_gpus:
         log_dict["perf/tokens_per_gpu_per_sec"] = sum(response_lengths) / rollout_time / args.rollout_num_gpus
     log_dict["perf/longest_sample_tokens_per_sec"] = max(response_lengths) / rollout_time
-    log_dict |= dict_add_prefix(_compute_metrics_from_samples(args, samples), f"rollout/")
+    log_dict |= dict_add_prefix(_compute_metrics_from_samples(args, samples), "rollout/")
     logger.info(f"perf {rollout_id}: {log_dict}")
     step = compute_rollout_step(args, rollout_id)
     log_dict["rollout/step"] = step
@@ -533,7 +533,7 @@ def _compute_metrics_from_samples(args, samples):
     response_lengths = [sample.effective_response_length for sample in samples]
 
     log_dict = {}
-    log_dict |= dict_add_prefix(compute_statistics(response_lengths), f"response_len/")
+    log_dict |= dict_add_prefix(compute_statistics(response_lengths), "response_len/")
     log_dict |= _compute_zero_std_metrics(args, samples)
     log_dict |= _compute_spec_metrics(args, samples)
     log_dict |= _compute_reward_cat_metrics(args, samples)
diff --git a/slime/ray/train_actor.py b/slime/ray/train_actor.py
@@ -81,7 +81,7 @@ def init(self, args, role, with_ref=False):
 
         try:
             if torch.version.hip is not None:
-                logger.info(f"Detected ROCm/HIP environment, skipping NUMA affinity setup")
+                logger.info("Detected ROCm/HIP environment, skipping NUMA affinity setup")
                 # will find the coresponding API to implement ROCm version as below
             else:
                 import pynvml
@@ -97,7 +97,7 @@ def init(self, args, role, with_ref=False):
                 pynvml.nvmlShutdown()
 
         except ImportError:
-            logger.info(f"Warning: pynvml not available, skipping NUMA affinity setup")
+            logger.info("Warning: pynvml not available, skipping NUMA affinity setup")
         except Exception as e:
             logger.info(f"Warning: Failed to set NUMA affinity: {e}")
 
diff --git a/slime/rollout/rm_hub/math_utils.py b/slime/rollout/rm_hub/math_utils.py
@@ -288,7 +288,7 @@ def _normalize(expr: str) -> str:
         "yard",
     ]:
         expr = re.sub(f"{unit}(es)?(s)? *(\^[0-9]+)?", "", expr)
-    expr = re.sub(f"\^ *\\\\circ", "", expr)
+    expr = re.sub("\^ *\\\\circ", "", expr)
 
     if len(expr) > 0 and expr[0] == "{" and expr[-1] == "}":
         expr = expr[1:-1]
diff --git a/slime/utils/arguments.py b/slime/utils/arguments.py
@@ -1550,7 +1550,7 @@ def hf_validate_args(args, hf_config):
                 )
 
     if len(errors) > 0:
-        raise AssertionError(f"hf_validate_args failed: " + "; ".join(errors))
+        raise AssertionError("hf_validate_args failed: " + "; ".join(errors))
 
 
 def _validate_and_update_megatron_args_from_hf(args, args_from_hf_config: Dict[str, Any]):
diff --git a/slime/utils/metric_checker.py b/slime/utils/metric_checker.py
@@ -26,4 +26,4 @@ def on_eval(self, metrics: Dict[str, float]):
 
     def dispose(self):
         assert self._exists_check_success, "[MetricChecker] accuracy check failed"
-        logger.info(f"[MetricChecker] pass dispose check")
+        logger.info("[MetricChecker] pass dispose check")
diff --git a/slime/utils/profile_utils.py b/slime/utils/profile_utils.py
@@ -130,7 +130,7 @@ def stop(self):
 class _MemrayMemoryProfiler(_BaseMemoryProfiler):
     def __init__(self, args):
         super().__init__(args)
-        assert args.memory_snapshot_num_steps is not None, f"In memray, must provide --memory-snapshot-num-steps"
+        assert args.memory_snapshot_num_steps is not None, "In memray, must provide --memory-snapshot-num-steps"
 
     def start(self):
         logger.info("Memray tracker started.")
diff --git a/tests/test_external_rollout.py b/tests/test_external_rollout.py
@@ -78,7 +78,7 @@ def execute():
         "--adam-beta2 0.98 "
     )
 
-    sglang_args = "--rollout-num-gpus-per-engine 2 " f"--sglang-mem-fraction-static 0.6 "
+    sglang_args = "--rollout-num-gpus-per-engine 2 " "--sglang-mem-fraction-static 0.6 "
 
     ci_args = (
         "--ci-test "
diff --git a/tests/test_quick_start_glm4_9B.py b/tests/test_quick_start_glm4_9B.py
@@ -30,7 +30,7 @@ def execute():
         "--num-rollout 3 "
         "--rollout-batch-size 8 "
         "--n-samples-per-prompt 8 "
-        f"--rollout-max-response-len 8192 "
+        "--rollout-max-response-len 8192 "
         "--rollout-temperature 0.8 "
         "--global-batch-size 32 "
         "--balance-data "
diff --git a/tools/convert_torch_dist_to_hf.py b/tools/convert_torch_dist_to_hf.py
@@ -73,7 +73,7 @@ def get_expert_param(args, name, param):
     if not match:
         assert param.shape[0] == num_experts
         for expert_id in range(num_experts):
-            expert_name = name.replace(".experts.experts.", f".experts.") + str(expert_id)
+            expert_name = name.replace(".experts.experts.", ".experts.") + str(expert_id)
             expert_param = param[expert_id]
             yield expert_name, expert_param
     else:

Original file line number	Diff line number	Diff line change
`@@ -134,7 +134,7 @@ def execute():`
`134`	`134`
`135`	`135`	`sglang_args = (`
`136`	`136`	`# "--rollout-num-gpus-per-engine 2 "`
`137`		`- f"--rollout-num-gpus-per-engine 8 " # temp use 1 engine per node to avoid flashinfer err`
	`137`	`+ "--rollout-num-gpus-per-engine 8 " # temp use 1 engine per node to avoid flashinfer err`
`138`	`138`	`"--sglang-mem-fraction-static 0.7 "`
`139`	`139`	`)`
`140`	`140`
Original file line number	Diff line number	Diff line change
`@@ -158,7 +158,7 @@ def train(args: ScriptArgs):`
`158`	`158`	`"--prompt-data /root/datasets/gsm8k/train.parquet "`
`159`	`159`	`"--input-key messages "`
`160`	`160`	`# Deliberately make it very short for this easy task`
`161`		`- f"--rollout-max-response-len 256 "`
	`161`	`+ "--rollout-max-response-len 256 "`
`162`	`162`	`)`
`163`	`163`	`eval_args += (`
`164`	`164`	`"--eval-prompt-data gsm8k /root/datasets/gsm8k/test.parquet "`