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": 76819.85838604625,
-      "required_pages_per_second_upper_estimate": 0.005314576049775728
+      "required_token_positions_per_second": 77384.34908444676,
+      "required_pages_per_second_upper_estimate": 0.005353628826089951
     },
     {
       "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": 9602.482298255782,
-      "required_pages_per_second_upper_estimate": 0.004020358815647119
+      "required_token_positions_per_second": 9673.043635555845,
+      "required_pages_per_second_upper_estimate": 0.004049901374086377
     },
     {
       "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": 19204.964596511563,
-      "required_pages_per_second_upper_estimate": 0.003999887656478385
+      "required_token_positions_per_second": 19346.08727111169,
+      "required_pages_per_second_upper_estimate": 0.004029279787942392
     },
     {
       "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": 28807.446894767345,
-      "required_pages_per_second_upper_estimate": 0.0017269624549167907
+      "required_token_positions_per_second": 29019.130906667535,
+      "required_pages_per_second_upper_estimate": 0.001739652588207436
     },
     {
       "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": 28807.446894767345,
-      "required_pages_per_second_upper_estimate": 0.0022396522855004566
+      "required_token_positions_per_second": 29019.130906667535,
+      "required_pages_per_second_upper_estimate": 0.0022561097863261297
     },
     {
       "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": 28807.446894767345,
-      "required_pages_per_second_upper_estimate": 0.3695109977394767
+      "required_token_positions_per_second": 29019.130906667535,
+      "required_pages_per_second_upper_estimate": 0.37222625295554873
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07531522333613339,
-    "producer_p95_seconds": 0.2879861610017542,
-    "gpu_update_seconds": 0.9597518343432945,
-    "producer_batches_per_second": 13.277528176966022,
-    "gpu_batches_per_second": 1.0419360132656013,
-    "producer_headroom": 12.74313202338792,
+    "producer_batch_seconds": 0.10117103117187298,
+    "producer_p95_seconds": 0.2711704980029026,
+    "gpu_update_seconds": 0.9527507935686489,
+    "producer_batches_per_second": 9.884252324177304,
+    "gpu_batches_per_second": 1.049592408371945,
+    "producer_headroom": 9.417229245692688,
     "sustainable": true,
-    "buffer_seconds": 60.46436556362755,
+    "buffer_seconds": 60.02329999482488,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 309296.6433266129,
-    "upload_bytes_per_second": 4861810.788286445,
+    "generation_bytes_per_second": 322491.0104688514,
+    "upload_bytes_per_second": 4817943.822503319,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.36785960053244,
+    "estimated_full_shards_per_hour": 69.1990636401096,
     "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": 1.590518836001138,
-      "bytes": 7289840,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 0.990031227000145,
       "bytes": 1215408,
@@ -562,6 +556,12 @@
       "bytes": 7607915,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.058052360000147,
+      "bytes": 885421,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -272,3 +272,4 @@
 {"step": 869700, "loss": 0.5524712800979614, "lr": 3e-05, "grad_norm": 0.3183505550061015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127299888487, "tokens_per_second": 152071.72078037952, "data_wait_s": 1.468299888074398e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 869800, "loss": 0.5738760828971863, "lr": 3e-05, "grad_norm": 0.3299766884051443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127314599732, "tokens_per_second": 152909.14793350216, "data_wait_s": 1.777400029823184e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 869900, "loss": 0.5243831872940063, "lr": 3e-05, "grad_norm": 0.3022764726004867, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127329234098, "tokens_per_second": 152506.0145587635, "data_wait_s": 1.138500010711141e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 870000, "loss": 0.4904373586177826, "lr": 3e-05, "grad_norm": 0.3192784582104852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127343745288, "tokens_per_second": 150244.0324024395, "data_wait_s": 9.524999768473208e-06, "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:ccc4415b6b0483a3d79a73ff95fb73b506c62b69f9a084027abbfe1a4210bc56
-size 584412
+oid sha256:87dedf67498d335749ba586d0bd3bfe141de493625ec2cd60e5b8779909b5939
+size 586530