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": 75518.54203847783,
-      "required_pages_per_second_upper_estimate": 0.006545755823900325
+      "required_token_positions_per_second": 75073.37973812624,
+      "required_pages_per_second_upper_estimate": 0.006507170283959385
     },
     {
       "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": 9439.817754809728,
-      "required_pages_per_second_upper_estimate": 0.003765684440246421
+      "required_token_positions_per_second": 9384.17246726578,
+      "required_pages_per_second_upper_estimate": 0.003743486703073951
     },
     {
       "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": 18879.635509619457,
-      "required_pages_per_second_upper_estimate": 0.00436747140491393
+      "required_token_positions_per_second": 18768.34493453156,
+      "required_pages_per_second_upper_estimate": 0.004341726289014571
     },
     {
       "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": 28319.453264429183,
-      "required_pages_per_second_upper_estimate": 0.0017245061231153794
+      "required_token_positions_per_second": 28152.517401797344,
+      "required_pages_per_second_upper_estimate": 0.0017143406049258828
     },
     {
       "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": 28319.453264429183,
-      "required_pages_per_second_upper_estimate": 0.0021801099055597864
+      "required_token_positions_per_second": 28152.517401797344,
+      "required_pages_per_second_upper_estimate": 0.0021672587207463437
     },
     {
       "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": 28319.453264429183,
-      "required_pages_per_second_upper_estimate": 0.35163283043109606
+      "required_token_positions_per_second": 28152.517401797344,
+      "required_pages_per_second_upper_estimate": 0.34956004571560084
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.20895707546111453,
-    "producer_p95_seconds": 0.3370537640003022,
-    "gpu_update_seconds": 0.9762900343393082,
-    "producer_batches_per_second": 4.785671879228101,
-    "gpu_batches_per_second": 1.0242857806868195,
-    "producer_headroom": 4.672203763308264,
+    "producer_batch_seconds": 0.09366953318749438,
+    "producer_p95_seconds": 0.306281631001184,
+    "gpu_update_seconds": 0.982079137201239,
+    "producer_batches_per_second": 10.675829866669046,
+    "gpu_batches_per_second": 1.0182478805626933,
+    "producer_headroom": 10.484509784365555,
     "sustainable": true,
-    "buffer_seconds": 60.52998212903711,
+    "buffer_seconds": 60.888906506476815,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 162,
+    "upload_pages_per_commit": 165,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 598700.3322615293,
+    "generation_bytes_per_second": 602099.8938201218,
     "upload_bytes_per_second": 2696283.522724811,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.23357713644222,
+    "estimated_full_shards_per_hour": 64.5983105228078,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -199,3 +199,4 @@
 {"step": 300100, "loss": 0.5722170490771532, "lr": 0.0003, "grad_norm": 0.1569018198689372, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43926970758, "tokens_per_second": 127397.35365565825, "data_wait_s": 1.1082996934419498e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 300200, "loss": 0.5093651041388512, "lr": 0.0003, "grad_norm": 0.13953843301649133, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43941504644, "tokens_per_second": 148812.316466676, "data_wait_s": 1.648900069994852e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 300300, "loss": 0.5240454412996769, "lr": 0.0003, "grad_norm": 0.16063872768604792, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43956154203, "tokens_per_second": 149920.6791598894, "data_wait_s": 1.0440002370160073e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 300400, "loss": 0.6158063635230064, "lr": 0.0003, "grad_norm": 0.14102001944295134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43970875888, "tokens_per_second": 149880.77992245305, "data_wait_s": 9.159000910585746e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f3c151144e91bc07ef4213159a31e2e65bc18dd28842d4443703a5cccb88059d
-size 429015
+oid sha256:68819768aed6dda7bb875c396e65fd873619580604d306dab7c089b62762d2fd
+size 431133