[Fix] Merge error fixes. (#152)

* PullRequest: 293 fix get_param_realloc_path

Merge branch xss/debug of git@code.alipay.com:inclusionAI/AReaL.git into gh
https://code.alipay.com/inclusionAI/AReaL/pull_requests/293

Reviewed-by: 博惟 <bowei.fw@antgroup.com>


* fix get_param_realloc_path

* PullRequest: 297 bugfix: reward is always -5

Merge branch xss/debug of git@code.alipay.com:inclusionAI/AReaL.git into gh
https://code.alipay.com/inclusionAI/AReaL/pull_requests/297

Reviewed-by: 博惟 <bowei.fw@antgroup.com>


* bugfix: reward is always -5

* PullRequest: 321 fix checkpoint save dir

Merge branch xss/debug of git@code.alipay.com:inclusionAI/AReaL.git into gh
https://code.alipay.com/inclusionAI/AReaL/pull_requests/321

Reviewed-by: 博惟 <bowei.fw@antgroup.com>


* fix checkpoint save dir

* PullRequest: 328 [Doc] update installation

Merge branch sxj/doc of git@code.alipay.com:inclusionAI/AReaL.git into gh
https://code.alipay.com/inclusionAI/AReaL/pull_requests/328

Reviewed-by: 博惟 <bowei.fw@antgroup.com>


* [Doc] update installation

---------

Co-authored-by: 温差 <xushusheng.xss@antgroup.com>
Co-authored-by: 冰临 <shenxujie.sxj@antgroup.com>

docs/tutorial/installation.md

@@ -67,6 +67,10 @@ cd AReaL
 bash examples/env/scripts/setup-pip-deps.sh
 ```
 
+::::{note}
+The SGLang patch is applied via `examples/env/scripts/setup-container-deps.sh` or `examples/env/scripts/setup-pip-deps.sh`. To confirm whether it has been applied, run `git status` in the `/sglang` directory (for Docker) or `AReaL/sglang` (for custom setups).
+::::
+
 ## (Optional) Launch Ray Cluster for Distributed Training
 
 On the first node, start the Ray Head:

realhf/base/constants.py

@@ -110,7 +110,7 @@ def get_save_path(args: "BaseExperimentConfig") -> str:
 
 
 def get_param_realloc_path(args: "BaseExperimentConfig"):
-    path = f"{args.cluster.fileroot}/.cache/{getpass.getuser()}/param_realloc"
+    path = f"{args.cluster.fileroot}/.cache/{getpass.getuser()}/param_realloc/{args.experiment_name}/{args.trial_name}"
     os.makedirs(path, exist_ok=True)
     return path
 

realhf/impl/dataset/math_parser.py

@@ -3,11 +3,11 @@
 import json
 import multiprocessing
 import re
-from concurrent.futures import ProcessPoolExecutor, as_completed
 from typing import List, Union
 
 import regex
 from latex2sympy2 import latex2sympy
+from pebble import ProcessPool
 from sympy import N, simplify
 from sympy.parsing.latex import parse_latex
 from sympy.parsing.sympy_parser import parse_expr
@@ -785,23 +785,6 @@ def process_results(answer, solution):
         return 0, ("None", "None")
 
 
-def process_results_process(a, b, output_queue):
-    result = process_results(a, b)
-    output_queue.put(result)
-
-
-def verify_math_solution(answer: str, solution: str):
-    # answer is generated by the model, solution is the ground truth
-    tmp = call_with_timeout(
-        process_results_process,
-        answer,
-        solution,
-    )
-    if isinstance(tmp, bool):
-        return 0
-    return tmp[0]
-
-
 def loadJson(dataDir):
     with open(dataDir, "r") as f:
         if dataDir.endswith(".jsonl"):
@@ -817,7 +800,7 @@ def parse_line(id2info, prompt_str, generated, query_id):
 
     label = 0
     for sol in info["solutions"]:
-        label = label or verify_math_solution(generated, sol)
+        label = label or process_results(generated, sol)
     return label
 
 
@@ -834,21 +817,26 @@ def parse_lines_in_parallel(
     )
 
     all_jobs = []
-    with ProcessPoolExecutor(max_workers=max_workers) as executor:
+    with ProcessPool(max_workers=max_workers) as executor:
         for qid, gen in zip(query_ids, generateds):
             info = id2info[qid.split("@idx:")[0]]
             jobs = []
             for sol in info["solutions"]:
-                job = executor.submit(verify_math_solution, gen, sol)
+                job = executor.schedule(process_results, args=[gen, sol], timeout=15)
                 jobs.append(job)
             all_jobs.append(jobs)
 
     labels = []
     for jobs in all_jobs:
         label = 0
-        for job in as_completed(jobs):
+        for job in jobs:
-            x = job.result()
+            try:
-            label = label or x
+                x = job.result()
+            except TimeoutError:
+                # print("[debug: timeout]")
+                logger.warning(f"Timeout occurred while justifying the math answer.")
+                x = (0, "timeout", "timeout")
+            label = label or x[0]
         labels.append(label)
     return labels
 

realhf/system/model_function_call.py

@@ -216,7 +216,7 @@ class ModelFunctionCall:
             for p in payloads.values():
                 p.post_hooks.append("save")
                 save_dir = os.path.join(
-                    constants.get_log_path(self.args),
+                    constants.get_save_path(self.args),
                     rpc.model_name.role,
                     f"epoch{ctrl.step_info.epoch + 1}"
                     f"epochstep{ctrl.step_info.epoch_step + 1}"