arbor-ai 0.1.9__py3-none-any.whl → 0.1.10__py3-none-any.whl

arbor/server/services/grpo_manager.py

@@ -2,6 +2,7 @@ import json
 import os
 import random
 import signal
+import socket
 import string
 import subprocess
 import sys
@@ -120,12 +121,17 @@ class GRPOManager:
 
         num_processes = self.settings.arbor_config.training.gpu_ids.count(",") + 1
 
+        # This is the port for the accelerate main process
+        main_process_port = get_free_port()
+
         params = [
             "python",
             "-m",
             "accelerate.commands.launch",
             "--num_processes",
             str(num_processes),
+            "--main_process_port",
+            str(main_process_port),
         ]
         if self.settings.arbor_config.training.accelerate_config:
             params.extend(
@@ -328,3 +334,12 @@ class GRPOManager:
         #     >= self.train_kwargs["update_interval"]
         # )
         return self.model_saved_and_reload_requested
+
+
+def get_free_port() -> int:
+    """
+    Return a free TCP port on localhost.
+    """
+    with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+        s.bind(("localhost", 0))
+        return s.getsockname()[1]

{arbor_ai-0.1.9.dist-info → arbor_ai-0.1.10.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: arbor-ai
-Version: 0.1.9
+Version: 0.1.10
 Summary: A framework for fine-tuning and managing language models
 Author-email: Noah Ziems <nziems2@nd.edu>
 Project-URL: Homepage, https://github.com/Ziems/arbor
@@ -57,6 +57,7 @@ inference:
 training:
   gpu_ids: '1, 2'
 ```
+Which will use the `GPU:0` for inference with `GPU:1` and `GPU:2` reserved for training. We generally recommend splitting the GPUs roughly evenly between inference and training.
 
 ### 2️⃣ Start the Server
 

{arbor_ai-0.1.9.dist-info → arbor_ai-0.1.10.dist-info}/RECORD

@@ -17,7 +17,7 @@ arbor/server/core/logging.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,
 arbor/server/services/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 arbor/server/services/dependencies.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 arbor/server/services/file_manager.py,sha256=Z9z4A4EzvPauid_DBfpim401DDtuJy_TbX4twTWDJWI,12119
-arbor/server/services/grpo_manager.py,sha256=Nd7T0q1RTQUhhG4uMB8lUonw_6Rww31yVJo8MM-noU8,12903
+arbor/server/services/grpo_manager.py,sha256=50g90lV8qpol7fQp2SBTXUCrF5eOP8YdxDnMLM0XY0E,13311
 arbor/server/services/inference_manager.py,sha256=gHI-Biy3TtGkyWxIDKY-uqZZm_fiQJLktkPY8ezRvo8,9660
 arbor/server/services/job_manager.py,sha256=m_d4UPwN_82f7t7K443DaFpFoyv7JZSZKml8tawt1Bk,2186
 arbor/server/services/training_manager.py,sha256=oQdhpfxdgp_lCTb_lxhvjupdLrcg6HL3TEbct_q9F6I,21065
@@ -26,9 +26,9 @@ arbor/server/services/comms/comms.py,sha256=3KN3mzwPvfW2_L5hq02JdAk6yOMyhY0_pBz-
 arbor/server/services/scripts/grpo_training.py,sha256=V36pCMZDJj2DdzquxScOddi9zP8EVPGWN3HGiftFfrY,21082
 arbor/server/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 arbor/server/utils/helpers.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-arbor_ai-0.1.9.dist-info/licenses/LICENSE,sha256=5vFGrbOFeXXM83JV9o16w7ohH4WLeu3-57GocJSz8ow,1067
-arbor_ai-0.1.9.dist-info/METADATA,sha256=GVTgqz290CXbwW5Yyt-EpelGmp1UZuK8kpN67Fp2bV0,2234
-arbor_ai-0.1.9.dist-info/WHEEL,sha256=0CuiUZ_p9E4cD6NyLD6UG80LBXYyiSYZOKDm5lp32xk,91
-arbor_ai-0.1.9.dist-info/entry_points.txt,sha256=PGBX-MfNwfIl8UPFgsX3gjtXLqSogRhOktKMpZUysD0,40
-arbor_ai-0.1.9.dist-info/top_level.txt,sha256=jzWdp3BRYqvZDMFsPajrcftvvlluzVDErkD8IMRfhYs,6
-arbor_ai-0.1.9.dist-info/RECORD,,
+arbor_ai-0.1.10.dist-info/licenses/LICENSE,sha256=5vFGrbOFeXXM83JV9o16w7ohH4WLeu3-57GocJSz8ow,1067
+arbor_ai-0.1.10.dist-info/METADATA,sha256=qnUBfdKczxenG5kPTcZgQVMnWimEUPExz7nONxBYpDQ,2413
+arbor_ai-0.1.10.dist-info/WHEEL,sha256=0CuiUZ_p9E4cD6NyLD6UG80LBXYyiSYZOKDm5lp32xk,91
+arbor_ai-0.1.10.dist-info/entry_points.txt,sha256=PGBX-MfNwfIl8UPFgsX3gjtXLqSogRhOktKMpZUysD0,40
+arbor_ai-0.1.10.dist-info/top_level.txt,sha256=jzWdp3BRYqvZDMFsPajrcftvvlluzVDErkD8IMRfhYs,6
+arbor_ai-0.1.10.dist-info/RECORD,,