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": 77173.78678960605,
-      "required_pages_per_second_upper_estimate": 0.005339061637961035
+      "required_token_positions_per_second": 76988.1525387257,
+      "required_pages_per_second_upper_estimate": 0.005326219029754327
     },
     {
       "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": 9646.723348700756,
-      "required_pages_per_second_upper_estimate": 0.004038881619610242
+      "required_token_positions_per_second": 9623.519067340712,
+      "required_pages_per_second_upper_estimate": 0.0040291664715652875
     },
     {
       "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": 19293.446697401512,
-      "required_pages_per_second_upper_estimate": 0.004018316144663844
+      "required_token_positions_per_second": 19247.038134681425,
+      "required_pages_per_second_upper_estimate": 0.004008650464925352
     },
     {
       "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": 28940.17004610227,
-      "required_pages_per_second_upper_estimate": 0.0017349190051828016
+      "required_token_positions_per_second": 28870.557202022137,
+      "required_pages_per_second_upper_estimate": 0.0017307458214728558
     },
     {
       "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": 28940.17004610227,
-      "required_pages_per_second_upper_estimate": 0.002249970926729301
+      "required_token_positions_per_second": 28870.557202022137,
+      "required_pages_per_second_upper_estimate": 0.002244558834296612
     },
     {
       "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": 28940.17004610227,
-      "required_pages_per_second_upper_estimate": 0.37121342781778416
+      "required_token_positions_per_second": 28870.557202022137,
+      "required_pages_per_second_upper_estimate": 0.37032050899837277
     }
   ],
   "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": 0.06926162537500291,
-    "producer_p95_seconds": 0.1268318909999948,
-    "gpu_update_seconds": 0.9553502953146503,
-    "producer_batches_per_second": 14.438009425648104,
-    "gpu_batches_per_second": 1.046736474468398,
-    "producer_headroom": 13.793356568548623,
+    "producer_batch_seconds": 0.07594268652343583,
+    "producer_p95_seconds": 0.12478036699997119,
+    "gpu_update_seconds": 0.957653841127233,
+    "producer_batches_per_second": 13.167824918748444,
+    "gpu_batches_per_second": 1.0442186487999905,
+    "producer_headroom": 12.610218112730342,
     "sustainable": true,
-    "buffer_seconds": 60.18706860482297,
+    "buffer_seconds": 60.33219199101568,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 61,
+    "upload_pages_per_commit": 91,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 1081743.9514613075,
-    "upload_bytes_per_second": 85005.02186575368,
-    "upload_sustainable": false,
+    "generation_bytes_per_second": 691027.1597500922,
+    "upload_bytes_per_second": 1783980.8016240546,
+    "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 116.05853513660153,
+    "estimated_full_shards_per_hour": 74.13917109669244,
     "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": [
@@ -190,6 +190,12 @@
       "bytes": 155991,
       "pages": 4,
       "shards": 1
+    },
+    {
+      "seconds": 0.9961135309999918,
+      "bytes": 5836820,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -1,2 +1,3 @@
 {"step": 604500, "loss": 0.636208713054657, "lr": 0.0001, "grad_norm": 0.16678688842682265, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88481063206, "tokens_per_second": 152754.29312491545, "data_wait_s": 1.4469000007011346e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 604600, "loss": 0.5752429366111755, "lr": 0.0001, "grad_norm": 0.17567589109079268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88495546086, "tokens_per_second": 149962.14570994265, "data_wait_s": 1.381399999900168e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 604700, "loss": 0.598007082939148, "lr": 0.0001, "grad_norm": 0.15348726900650103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88510111031, "tokens_per_second": 150957.2267887229, "data_wait_s": 1.3324000065040309e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8256db8c9b4f8c98e7a0714b0a352d6d3323c7ee1fec0a9f707d5990f51886f2
-size 4401
+oid sha256:a5c6c742e4cd174071fef0481326f3cb1fb950c966a39350830a73f4f5ace701
+size 6519