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": 75405.72835431593,
-      "required_pages_per_second_upper_estimate": 0.0065359774196808405
+      "required_token_positions_per_second": 75589.5095836848,
+      "required_pages_per_second_upper_estimate": 0.006551907110853268
     },
     {
       "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": 9425.716044289491,
-      "required_pages_per_second_upper_estimate": 0.0037600590570805374
+      "required_token_positions_per_second": 9448.6886979606,
+      "required_pages_per_second_upper_estimate": 0.0037692231921017233
     },
     {
       "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": 18851.432088578982,
-      "required_pages_per_second_upper_estimate": 0.004360947039819474
+      "required_token_positions_per_second": 18897.3773959212,
+      "required_pages_per_second_upper_estimate": 0.004371575678063304
     },
     {
       "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": 28277.148132868475,
-      "required_pages_per_second_upper_estimate": 0.0017219299625612038
+      "required_token_positions_per_second": 28346.066093881804,
+      "required_pages_per_second_upper_estimate": 0.0017261267048023189
     },
     {
       "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": 28277.148132868475,
-      "required_pages_per_second_upper_estimate": 0.002176853139424144
+      "required_token_positions_per_second": 28346.066093881804,
+      "required_pages_per_second_upper_estimate": 0.0021821586348400764
     },
     {
       "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": 28277.148132868475,
-      "required_pages_per_second_upper_estimate": 0.35110754228327945
+      "required_token_positions_per_second": 28346.066093881804,
+      "required_pages_per_second_upper_estimate": 0.35196327270548694
     }
   ],
   "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": 0.04086105484377356,
-    "producer_p95_seconds": 0.13841988099920854,
-    "gpu_update_seconds": 0.9777506511649535,
-    "producer_batches_per_second": 24.473181219216148,
-    "gpu_batches_per_second": 1.0227556471668284,
-    "producer_headroom": 23.928668873166497,
+    "producer_batch_seconds": 0.03816816335941553,
+    "producer_p95_seconds": 0.1314340840017394,
+    "gpu_update_seconds": 0.9753734401249959,
+    "producer_batches_per_second": 26.199845944468652,
+    "gpu_batches_per_second": 1.025248339622461,
+    "producer_headroom": 25.554633869601314,
     "sustainable": true,
-    "buffer_seconds": 60.62054037222712,
+    "buffer_seconds": 60.473153287749746,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 69,
+    "upload_pages_per_commit": 72,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -15,3 +15,4 @@
 {"step": 86500, "loss": 0.6315119042992592, "lr": 0.0003, "grad_norm": 0.13389635879207812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12658762758, "tokens_per_second": 150075.29958827473, "data_wait_s": 8.196999260690063e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 86600, "loss": 0.6253555119037628, "lr": 0.0003, "grad_norm": 0.1284965926247556, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12673432090, "tokens_per_second": 149705.77500699987, "data_wait_s": 8.79099934536498e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 86700, "loss": 0.6090225968509912, "lr": 0.0003, "grad_norm": 0.13672678645286673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12687942064, "tokens_per_second": 148855.0959608192, "data_wait_s": 8.787999831838533e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 86800, "loss": 0.6323533430695534, "lr": 0.0003, "grad_norm": 0.13809571203695895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12702535251, "tokens_per_second": 149576.5669256539, "data_wait_s": 8.88800059328787e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:069539355107a0c45092eece84b635aca692bc86945daae0de632e234d2a4f72
-size 36171
+oid sha256:c40117c64661cfb0d60ae4dab9cdd6d5270702243a085861664b0b42d20c18ea
+size 38289