Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77360.33365193554,
-      "required_pages_per_second_upper_estimate": 0.005351967382745344
+      "required_token_positions_per_second": 77325.40081817229,
+      "required_pages_per_second_upper_estimate": 0.0053495506482502606
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9670.041706491942,
-      "required_pages_per_second_upper_estimate": 0.004048644529074728
+      "required_token_positions_per_second": 9665.675102271536,
+      "required_pages_per_second_upper_estimate": 0.004046816323072709
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19340.083412983884,
-      "required_pages_per_second_upper_estimate": 0.004028029342626756
+      "required_token_positions_per_second": 19331.350204543072,
+      "required_pages_per_second_upper_estimate": 0.004026210445618729
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29010.125119475826,
-      "required_pages_per_second_upper_estimate": 0.0017391127050163338
+      "required_token_positions_per_second": 28997.025306814605,
+      "required_pages_per_second_upper_estimate": 0.0017383273912495495
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29010.125119475826,
-      "required_pages_per_second_upper_estimate": 0.0022554096259842457
+      "required_token_positions_per_second": 28997.025306814605,
+      "required_pages_per_second_upper_estimate": 0.0022543911731698224
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29010.125119475826,
-      "required_pages_per_second_upper_estimate": 0.3721107363871144
+      "required_token_positions_per_second": 28997.025306814605,
+      "required_pages_per_second_upper_estimate": 0.3719427060557792
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.047578922648453315,
-    "producer_p95_seconds": 0.13182983900014733,
-    "gpu_update_seconds": 0.9530465617136766,
-    "producer_batches_per_second": 21.01771003493934,
-    "gpu_batches_per_second": 1.0492666782217819,
-    "producer_headroom": 20.03085628389398,
+    "producer_batch_seconds": 0.05282973203907737,
+    "producer_p95_seconds": 0.13105633600002875,
+    "gpu_update_seconds": 0.9534771138576904,
+    "producer_batches_per_second": 18.92873503996414,
+    "gpu_batches_per_second": 1.0487928713402273,
+    "producer_headroom": 18.048115654881943,
     "sustainable": true,
-    "buffer_seconds": 60.04193338796163,
+    "buffer_seconds": 60.0690581730345,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 62,
+    "upload_pages_per_commit": 60,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791033297667653750-ba4975eb/train.jsonl

@@ -7,3 +7,4 @@
 {"step": 600700, "loss": 0.5805760025978088, "lr": 0.0003, "grad_norm": 0.19626544568017945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87924785186, "tokens_per_second": 156109.97371656948, "data_wait_s": 1.483399978496891e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 600800, "loss": 0.58770352602005, "lr": 0.0003, "grad_norm": 0.14629567278331893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87939498864, "tokens_per_second": 154325.5528206508, "data_wait_s": 1.5793999864399666e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 600900, "loss": 0.5725371837615967, "lr": 0.0003, "grad_norm": 0.16399186990356174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87954036705, "tokens_per_second": 152405.63047677095, "data_wait_s": 1.699000017651997e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 601000, "loss": 0.6038609147071838, "lr": 0.0003, "grad_norm": 0.14563901031957593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87968610536, "tokens_per_second": 152654.9884336665, "data_wait_s": 1.4574000033462653e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791033297.83ac8725-d818-41af-bbb1-78dcd88792cf-7zwhs.737.0.1791033297667653750-ba4975eb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:27703eb7af2ed35f1734200fd4737f8014479e601223076ffe2df5659705009d
-size 19227
+oid sha256:50b3dab3e5557f62124a41bddb028b098fb41f2e99b24f152e560fb199a31a64
+size 21345