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": 74012.55562262743,
-      "required_pages_per_second_upper_estimate": 0.006415220738262089
+      "required_token_positions_per_second": 75314.24194785656,
+      "required_pages_per_second_upper_estimate": 0.00652804760983912
     },
     {
       "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": 9251.56945282843,
-      "required_pages_per_second_upper_estimate": 0.0036905893780231486
+      "required_token_positions_per_second": 9414.28024348207,
+      "required_pages_per_second_upper_estimate": 0.003755497145157998
     },
     {
       "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": 18503.13890565686,
-      "required_pages_per_second_upper_estimate": 0.004280375541701103
+      "required_token_positions_per_second": 18828.56048696414,
+      "required_pages_per_second_upper_estimate": 0.00435565609731232
     },
     {
       "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": 27754.708358485288,
-      "required_pages_per_second_upper_estimate": 0.001690116121330926
+      "required_token_positions_per_second": 28242.84073044621,
+      "required_pages_per_second_upper_estimate": 0.0017198408217507174
     },
     {
       "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": 27754.708358485288,
-      "required_pages_per_second_upper_estimate": 0.0021366342793862694
+      "required_token_positions_per_second": 28242.84073044621,
+      "required_pages_per_second_upper_estimate": 0.0021742120606167096
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27754.708358485288,
-      "required_pages_per_second_upper_estimate": 0.3446205887788878
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28242.84073044621,
+      "required_pages_per_second_upper_estimate": 0.3506815591646847
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2632568077815165,
-    "producer_p95_seconds": 0.305773735002731,
-    "gpu_update_seconds": 0.9961553060797101,
-    "producer_batches_per_second": 3.7985722322893367,
-    "gpu_batches_per_second": 1.0038595326419737,
-    "producer_headroom": 3.783967884722072,
+    "producer_batch_seconds": 0.0945571504531415,
+    "producer_p95_seconds": 0.27301258300212794,
+    "gpu_update_seconds": 0.9789383533999482,
+    "producer_batches_per_second": 10.575614802347046,
+    "gpu_batches_per_second": 1.0215147833639397,
+    "producer_headroom": 10.352874940801735,
     "sustainable": true,
-    "buffer_seconds": 60.76547367086232,
+    "buffer_seconds": 60.69417791079679,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 454662.3409855517,
-    "upload_bytes_per_second": 2845317.0740763033,
+    "generation_bytes_per_second": 457089.45913459087,
+    "upload_bytes_per_second": 2856798.250413766,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.55995437788881,
+    "estimated_full_shards_per_hour": 98.08075743225379,
     "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": 0.9817681739987165,
-      "bytes": 8356010,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.2058971119986381,
       "bytes": 10431329,
@@ -562,6 +556,12 @@
       "bytes": 2356968,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.1747013890017115,
+      "bytes": 10632168,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -237,3 +237,4 @@
 {"step": 218700, "loss": 0.5589394718408585, "lr": 0.0003, "grad_norm": 0.1404609947939712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32010632638, "tokens_per_second": 147477.66888674113, "data_wait_s": 1.5420999261550605e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 218800, "loss": 0.6002527214586735, "lr": 0.0003, "grad_norm": 0.16279575948199368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32025062644, "tokens_per_second": 146519.64073062586, "data_wait_s": 1.3880999176762998e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 218900, "loss": 0.5877895951271057, "lr": 0.0003, "grad_norm": 0.16471820727193348, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32039711182, "tokens_per_second": 149834.3093207871, "data_wait_s": 1.4344997907755896e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 219000, "loss": 0.635651670396328, "lr": 0.0003, "grad_norm": 0.1789175456113893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32054448077, "tokens_per_second": 149585.77841550292, "data_wait_s": 1.5058998542372137e-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 sha256:50a1a1c32b6c5b698fd31f5b3dd3131a5c4c472d177cd0230f3e907c87c36472
-size 509499
+oid sha256:ddfe014040b28ed4146fa733f2d45cbffcc7cd3126c4d8fc3449d8bd3a14ec42
+size 511617