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": 75295.43667913652,
-      "required_pages_per_second_upper_estimate": 0.0065264176168079845
+      "required_token_positions_per_second": 76028.65767843553,
+      "required_pages_per_second_upper_estimate": 0.00658997135469559
     },
     {
       "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": 9411.929584892065,
-      "required_pages_per_second_upper_estimate": 0.0037545594323009674
+      "required_token_positions_per_second": 9503.582209804441,
+      "required_pages_per_second_upper_estimate": 0.003791121034707372
     },
     {
       "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": 18823.85916978413,
-      "required_pages_per_second_upper_estimate": 0.004354568530323081
+      "required_token_positions_per_second": 19007.164419608882,
+      "required_pages_per_second_upper_estimate": 0.004396972973807824
     },
     {
       "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": 28235.788754676196,
-      "required_pages_per_second_upper_estimate": 0.001719411393425181
+      "required_token_positions_per_second": 28510.746629413323,
+      "required_pages_per_second_upper_estimate": 0.0017361548854041928
     },
     {
       "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": 28235.788754676196,
-      "required_pages_per_second_upper_estimate": 0.0021736691800008155
+      "required_token_positions_per_second": 28510.746629413323,
+      "required_pages_per_second_upper_estimate": 0.0021948361983301916
     },
     {
       "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": 28235.788754676196,
-      "required_pages_per_second_upper_estimate": 0.35059399722706575
+      "required_token_positions_per_second": 28510.746629413323,
+      "required_pages_per_second_upper_estimate": 0.3540080538064905
     }
   ],
   "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": 0.12151216214834903,
-    "producer_p95_seconds": 0.2743932310004311,
-    "gpu_update_seconds": 0.9791828462883351,
-    "producer_batches_per_second": 8.22962888915714,
-    "gpu_batches_per_second": 1.0212597205829064,
-    "producer_headroom": 8.058311439581598,
+    "producer_batch_seconds": 0.2024880375624889,
+    "producer_p95_seconds": 0.19761140199989313,
+    "gpu_update_seconds": 0.9697395988738062,
+    "producer_batches_per_second": 4.938563344471125,
+    "gpu_batches_per_second": 1.031204666862461,
+    "producer_headroom": 4.789120436680312,
     "sustainable": true,
-    "buffer_seconds": 60.709336469876774,
+    "buffer_seconds": 60.12385513017598,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 415297.4952526611,
-    "upload_bytes_per_second": 5184286.961140318,
+    "generation_bytes_per_second": 402295.39350289333,
+    "upload_bytes_per_second": 5216543.443157296,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.11317485031962,
+    "estimated_full_shards_per_hour": 86.3232264882574,
     "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": 0.7755766290001702,
-      "bytes": 1144894,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.4222684020005545,
       "bytes": 6265294,
@@ -562,6 +556,12 @@
       "bytes": 2711632,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.13261040100042,
+      "bytes": 5584296,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -121,3 +121,4 @@
 {"step": 167100, "loss": 0.641258591786027, "lr": 0.0003, "grad_norm": 0.16054336133288577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24455944115, "tokens_per_second": 150415.35592970694, "data_wait_s": 1.2121001418563537e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 167200, "loss": 0.5853394120931625, "lr": 0.0003, "grad_norm": 0.14848904564804077, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24470597460, "tokens_per_second": 149105.21808988327, "data_wait_s": 8.761999197304249e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 167300, "loss": 0.6270021367818117, "lr": 0.0003, "grad_norm": 0.12998750910320842, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24485168322, "tokens_per_second": 149277.00870477536, "data_wait_s": 1.4106999515206553e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 167400, "loss": 0.5898058451712132, "lr": 0.0003, "grad_norm": 0.13277469101355197, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24499680534, "tokens_per_second": 148153.1811235543, "data_wait_s": 1.4287999874795787e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:76774aa820defe041bf372748f255d1ded869dfbc763435c1c853362f095a113
-size 262245
+oid sha256:45383e5a83a362ce5bb3840766de420c1f4d6097d6de5c1beed9b5d5887761e2
+size 264363