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": 76577.90463803377,
-      "required_pages_per_second_upper_estimate": 0.005297837128078182
+      "required_token_positions_per_second": 77284.56310046758,
+      "required_pages_per_second_upper_estimate": 0.0053467254001828895
     },
     {
       "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": 9572.238079754221,
-      "required_pages_per_second_upper_estimate": 0.004007696192931617
+      "required_token_positions_per_second": 9660.570387558448,
+      "required_pages_per_second_upper_estimate": 0.004044679085620904
     },
     {
       "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": 19144.476159508442,
-      "required_pages_per_second_upper_estimate": 0.0039872895102379905
+      "required_token_positions_per_second": 19321.140775116895,
+      "required_pages_per_second_upper_estimate": 0.004024084090710017
     },
     {
       "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": 28716.714239262663,
-      "required_pages_per_second_upper_estimate": 0.001721523170760028
+      "required_token_positions_per_second": 28981.711162675343,
+      "required_pages_per_second_upper_estimate": 0.001737409331691745
     },
     {
       "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": 28716.714239262663,
-      "required_pages_per_second_upper_estimate": 0.0022325982206257418
+      "required_token_positions_per_second": 28981.711162675343,
+      "required_pages_per_second_upper_estimate": 0.002253200565822796
     },
     {
       "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": 28716.714239262663,
-      "required_pages_per_second_upper_estimate": 0.3683471766557979
+      "required_token_positions_per_second": 28981.711162675343,
+      "required_pages_per_second_upper_estimate": 0.3717462726578077
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.40197195971875743,
-    "producer_p95_seconds": 0.2049835260004329,
-    "gpu_update_seconds": 0.962784243686157,
-    "producer_batches_per_second": 2.4877357134553795,
-    "gpu_batches_per_second": 1.0386543055288868,
-    "producer_headroom": 2.3951527473701795,
+    "producer_batch_seconds": 0.06899631614844992,
+    "producer_p95_seconds": 0.1833469649991457,
+    "gpu_update_seconds": 0.9539809380064146,
+    "producer_batches_per_second": 14.493527420339905,
+    "gpu_batches_per_second": 1.0482389743444496,
+    "producer_headroom": 13.826548883477555,
     "sustainable": true,
-    "buffer_seconds": 60.65540735222789,
+    "buffer_seconds": 60.10079909440412,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 335950.1157080666,
-    "upload_bytes_per_second": 4550117.712570053,
+    "generation_bytes_per_second": 354656.5022857757,
+    "upload_bytes_per_second": 4533879.065807922,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.08707431251048,
+    "estimated_full_shards_per_hour": 76.10102940969423,
     "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": [
@@ -544,6 +544,18 @@
       "bytes": 6214888,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.4851975340006902,
+      "bytes": 6638444,
+      "pages": 30,
+      "shards": 3
+    },
+    {
+      "seconds": 1.5820424499997898,
+      "bytes": 5864908,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -73,3 +73,4 @@
 {"step": 849800, "loss": 0.5476540923118591, "lr": 3e-05, "grad_norm": 0.2951122253907351, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124387247556, "tokens_per_second": 151878.84600292164, "data_wait_s": 1.462500040361192e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 849900, "loss": 0.5229765772819519, "lr": 3e-05, "grad_norm": 0.3029686773333047, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124401978036, "tokens_per_second": 141574.2633784317, "data_wait_s": 1.4706000001751818e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 850000, "loss": 0.5116078853607178, "lr": 3e-05, "grad_norm": 0.2902432474844378, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124416666044, "tokens_per_second": 152880.71914155624, "data_wait_s": 7.378999725915492e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 850100, "loss": 0.563275933265686, "lr": 3e-05, "grad_norm": 0.30711087724013664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124431441815, "tokens_per_second": 146640.6605470716, "data_wait_s": 1.2968000191904139e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3ed761234375a8734d234276cc5caab3b39a66e43893985b107a0f93aedc2efd
-size 160581
+oid sha256:6fe8a11f85161e488f07a44cd779b8acd70b9f071a30addcb8ffd49917d925ef
+size 162699