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": 75713.87319249971,
-      "required_pages_per_second_upper_estimate": 0.006562686633268659
+      "required_token_positions_per_second": 75535.3813282497,
+      "required_pages_per_second_upper_estimate": 0.006547215410858962
     },
     {
       "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": 9464.234149062464,
-      "required_pages_per_second_upper_estimate": 0.0037754245049714634
+      "required_token_positions_per_second": 9441.922666031212,
+      "required_pages_per_second_upper_estimate": 0.0037665241208038984
     },
     {
       "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": 18928.468298124928,
-      "required_pages_per_second_upper_estimate": 0.004378768011106946
+      "required_token_positions_per_second": 18883.845332062425,
+      "required_pages_per_second_upper_estimate": 0.004368445273245766
     },
     {
       "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": 28392.702447187392,
-      "required_pages_per_second_upper_estimate": 0.0017289666140365935
+      "required_token_positions_per_second": 28325.767998093637,
+      "required_pages_per_second_upper_estimate": 0.0017248906572646977
     },
     {
       "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": 28392.702447187392,
-      "required_pages_per_second_upper_estimate": 0.002185748830415229
+      "required_token_positions_per_second": 28325.767998093637,
+      "required_pages_per_second_upper_estimate": 0.002180596031237578
     },
     {
       "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": 28392.702447187392,
-      "required_pages_per_second_upper_estimate": 0.3525423401317083
+      "required_token_positions_per_second": 28325.767998093637,
+      "required_pages_per_second_upper_estimate": 0.3517112382891545
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13291322080476675,
-    "producer_p95_seconds": 0.36993682799948147,
-    "gpu_update_seconds": 0.9737713432325579,
-    "producer_batches_per_second": 7.523706023713604,
-    "gpu_batches_per_second": 1.0269351290215347,
-    "producer_headroom": 7.3263693207984835,
+    "producer_batch_seconds": 0.11099471561712448,
+    "producer_p95_seconds": 0.3792780860021594,
+    "gpu_update_seconds": 0.9760723875822448,
+    "producer_batches_per_second": 9.009437921797044,
+    "gpu_batches_per_second": 1.0245141781717897,
+    "producer_headroom": 8.793863583102459,
     "sustainable": true,
-    "buffer_seconds": 60.37382328041859,
+    "buffer_seconds": 60.516488030099175,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 270031.1691939612,
-    "upload_bytes_per_second": 5291042.556009083,
+    "generation_bytes_per_second": 297295.53941490006,
+    "upload_bytes_per_second": 5315379.803461301,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.94240290512206,
+    "estimated_full_shards_per_hour": 63.79270207247974,
     "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": [
-    {
-      "seconds": 0.9600440670001262,
-      "bytes": 1151300,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.3565457900003821,
       "bytes": 7973344,
@@ -562,6 +556,12 @@
       "bytes": 5919458,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.8879908719973173,
+      "bytes": 2645607,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -314,3 +314,4 @@
 {"step": 226400, "loss": 0.5998847484588623, "lr": 0.0003, "grad_norm": 0.14292261352133273, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33138742060, "tokens_per_second": 147449.55271050605, "data_wait_s": 2.0502997358562425e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 226500, "loss": 0.6257619615644217, "lr": 0.0003, "grad_norm": 0.15677036568711947, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33153327328, "tokens_per_second": 148919.01146690917, "data_wait_s": 9.450999641558155e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 226600, "loss": 0.5676675904542208, "lr": 0.0003, "grad_norm": 0.13611202505926445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33168023130, "tokens_per_second": 148936.78350129872, "data_wait_s": 8.999999408842996e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 226700, "loss": 0.6044607497751713, "lr": 0.0003, "grad_norm": 0.14674524639210093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33182695458, "tokens_per_second": 148143.8770490476, "data_wait_s": 1.906300167320296e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d5c4ac8c5c4b6cbd63b3c629aaa45d2c59d95830b3b3fb208681b0d9b9565746
-size 674151
+oid sha256:c105fc0e21f56d0363e505a5d1c515ea1eef6652841cc41d1d17a57e6a6a7ac3
+size 676269