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": 75700.24956836377,
-      "required_pages_per_second_upper_estimate": 0.006561505772057314
+      "required_token_positions_per_second": 75638.8041231787,
+      "required_pages_per_second_upper_estimate": 0.00655617983660073
     },
     {
       "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": 9462.531196045471,
-      "required_pages_per_second_upper_estimate": 0.003774745171551568
+      "required_token_positions_per_second": 9454.850515397338,
+      "required_pages_per_second_upper_estimate": 0.0037716812332046186
     },
     {
       "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": 18925.062392090942,
-      "required_pages_per_second_upper_estimate": 0.0043779801146878825
+      "required_token_positions_per_second": 18909.701030794677,
+      "required_pages_per_second_upper_estimate": 0.004374426534102717
     },
     {
       "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": 28387.593588136413,
-      "required_pages_per_second_upper_estimate": 0.0017286555113245007
+      "required_token_positions_per_second": 28364.551546192015,
+      "required_pages_per_second_upper_estimate": 0.0017272523718623379
     },
     {
       "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": 28387.593588136413,
-      "required_pages_per_second_upper_estimate": 0.002185355536303258
+      "required_token_positions_per_second": 28364.551546192015,
+      "required_pages_per_second_upper_estimate": 0.002183581696129923
     },
     {
       "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": 28387.593588136413,
-      "required_pages_per_second_upper_estimate": 0.3524789052005465
+      "required_token_positions_per_second": 28364.551546192015,
+      "required_pages_per_second_upper_estimate": 0.3521928001563507
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.08760824885148111,
-    "producer_p95_seconds": 0.21248604699940188,
-    "gpu_update_seconds": 0.9739465909345165,
-    "producer_batches_per_second": 11.414450272773532,
-    "gpu_batches_per_second": 1.0267503467931285,
-    "producer_headroom": 11.117064930559344,
+    "producer_batch_seconds": 0.10962028046090211,
+    "producer_p95_seconds": 0.2130781380001281,
+    "gpu_update_seconds": 0.9747377798296898,
+    "producer_batches_per_second": 9.122399576022492,
+    "gpu_batches_per_second": 1.0259169396047458,
+    "producer_headroom": 8.891947509451466,
     "sustainable": true,
-    "buffer_seconds": 60.38468863794002,
+    "buffer_seconds": 60.43374234944077,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 239527.41489859487,
-    "upload_bytes_per_second": 5752375.986722912,
+    "generation_bytes_per_second": 249185.21753235153,
+    "upload_bytes_per_second": 5838934.00826427,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 51.39700732439408,
+    "estimated_full_shards_per_hour": 53.46934694745931,
     "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": 0.9485998250002012,
-      "bytes": 1124293,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.2218557999995028,
       "bytes": 7398978,
@@ -562,6 +556,12 @@
       "bytes": 2694003,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.6696938850000151,
+      "bytes": 11635327,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -134,3 +134,4 @@
 {"step": 368400, "loss": 0.5785097442567348, "lr": 0.0003, "grad_norm": 0.1597017862073903, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53923384418, "tokens_per_second": 148113.3441642095, "data_wait_s": 1.0154999472433701e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 368500, "loss": 0.5769478399306536, "lr": 0.0003, "grad_norm": 0.16911262658034257, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53938125446, "tokens_per_second": 150158.0008052164, "data_wait_s": 1.0830000974237919e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 368600, "loss": 0.5898146387189627, "lr": 0.0003, "grad_norm": 0.1585002514120512, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53952706554, "tokens_per_second": 148790.51250604712, "data_wait_s": 1.0783000107039697e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 368700, "loss": 0.562000397592783, "lr": 0.0003, "grad_norm": 0.14543126631900252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53967315924, "tokens_per_second": 148058.31468745435, "data_wait_s": 1.321800118603278e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3d1e658589251cf65e9f27557631f1b95ce32f12bda26d5ac64d13b4df55aae4
-size 289779
+oid sha256:fa391a595c96df643b2b03658f80dff17be9328d6eb398e1f89881f88df3d139
+size 291897