Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75847.45591587861,
-      "required_pages_per_second_upper_estimate": 0.006574265245168825
+      "required_token_positions_per_second": 75953.55001164008,
+      "required_pages_per_second_upper_estimate": 0.006583461212496401
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9480.931989484827,
-      "required_pages_per_second_upper_estimate": 0.003782085523170906
+      "required_token_positions_per_second": 9494.19375145501,
+      "required_pages_per_second_upper_estimate": 0.003787375838301823
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18961.863978969654,
-      "required_pages_per_second_upper_estimate": 0.0043864935141264645
+      "required_token_positions_per_second": 18988.38750291002,
+      "required_pages_per_second_upper_estimate": 0.00439262926459199
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28442.795968454484,
-      "required_pages_per_second_upper_estimate": 0.001732017046661365
+      "required_token_positions_per_second": 28482.581254365028,
+      "required_pages_per_second_upper_estimate": 0.001734439761836054
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28442.795968454484,
-      "required_pages_per_second_upper_estimate": 0.002189605168350112
+      "required_token_positions_per_second": 28482.581254365028,
+      "required_pages_per_second_upper_estimate": 0.0021926679497922255
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28442.795968454484,
-      "required_pages_per_second_upper_estimate": 0.35316433401361463
+      "required_token_positions_per_second": 28482.581254365028,
+      "required_pages_per_second_upper_estimate": 0.35365833411183717
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.0541209528437534,
-    "producer_p95_seconds": 0.13950678800006244,
-    "gpu_update_seconds": 0.9720563347803085,
-    "producer_batches_per_second": 18.477132191057112,
-    "gpu_batches_per_second": 1.0287469606645863,
-    "producer_headroom": 17.96081339489023,
+    "producer_batch_seconds": 0.048399034882811165,
+    "producer_p95_seconds": 0.1318859010000324,
+    "gpu_update_seconds": 0.9706985386292147,
+    "producer_batches_per_second": 20.6615690255251,
+    "gpu_batches_per_second": 1.0301859539339204,
+    "producer_headroom": 20.056154858863863,
     "sustainable": true,
-    "buffer_seconds": 60.26749275637913,
+    "buffer_seconds": 60.183309395011314,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 64,
+    "upload_pages_per_commit": 67,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -12,3 +12,4 @@
 {"step": 156200, "loss": 0.576857328414917, "lr": 0.0003, "grad_norm": 0.13306192544028528, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22859969626, "tokens_per_second": 150264.58248178678, "data_wait_s": 1.6371000128856394e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 156300, "loss": 0.5939075406640768, "lr": 0.0003, "grad_norm": 0.15584521523566924, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22874747073, "tokens_per_second": 151883.96645317494, "data_wait_s": 1.097200015465205e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 156400, "loss": 0.5991663802415133, "lr": 0.0003, "grad_norm": 0.14258397599734335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22889448533, "tokens_per_second": 151256.05648356315, "data_wait_s": 1.3411000054475153e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 156500, "loss": 0.6319344639778137, "lr": 0.0003, "grad_norm": 0.14753616763741959, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22904209759, "tokens_per_second": 151507.8120863896, "data_wait_s": 1.3636999938171357e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4b382c2b90c52b67d1b17711d3505337384d558ad5e09663b69c79afafc2db2e
-size 29817
+oid sha256:796f297baa092086e312b5752309616dcf74304dbb855c6b3d7d1f10d07d5cf8
+size 31935