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": 75436.78399252897,
-      "required_pages_per_second_upper_estimate": 0.006538669243691351
+      "required_token_positions_per_second": 75437.46000741117,
+      "required_pages_per_second_upper_estimate": 0.006538727838947996
     },
     {
       "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": 9429.597999066122,
-      "required_pages_per_second_upper_estimate": 0.0037616076268813313
+      "required_token_positions_per_second": 9429.682500926396,
+      "required_pages_per_second_upper_estimate": 0.00376164133593681
     },
     {
       "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": 18859.195998132243,
-      "required_pages_per_second_upper_estimate": 0.004362743083654479
+      "required_token_positions_per_second": 18859.36500185279,
+      "required_pages_per_second_upper_estimate": 0.004362782179690864
     },
     {
       "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": 28288.79399719836,
-      "required_pages_per_second_upper_estimate": 0.0017226391345977659
+      "required_token_positions_per_second": 28289.047502779184,
+      "required_pages_per_second_upper_estimate": 0.001722654571757597
     },
     {
       "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": 28288.79399719836,
-      "required_pages_per_second_upper_estimate": 0.0021777496702980745
+      "required_token_positions_per_second": 28289.047502779184,
+      "required_pages_per_second_upper_estimate": 0.002177769185859434
     },
     {
       "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": 28288.79399719836,
-      "required_pages_per_second_upper_estimate": 0.351252144942056
+      "required_token_positions_per_second": 28289.047502779184,
+      "required_pages_per_second_upper_estimate": 0.3512552926329412
     }
   ],
   "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": 0.7087429197030986,
-    "producer_p95_seconds": 0.3774812919982651,
-    "gpu_update_seconds": 0.9773481330712852,
-    "producer_batches_per_second": 1.4109488394168548,
-    "gpu_batches_per_second": 1.023176866218112,
-    "producer_headroom": 1.3789882140631595,
+    "producer_batch_seconds": 0.1250306849065339,
+    "producer_p95_seconds": 0.3667360049985291,
+    "gpu_update_seconds": 0.9773393747981014,
+    "producer_batches_per_second": 7.998036647944025,
+    "gpu_batches_per_second": 1.0231860352567703,
+    "producer_headroom": 7.816796137113916,
     "sustainable": true,
-    "buffer_seconds": 60.595584250419684,
+    "buffer_seconds": 60.595041237482285,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 303286.03470474394,
-    "upload_bytes_per_second": 4511343.990475607,
+    "generation_bytes_per_second": 304021.63174405484,
+    "upload_bytes_per_second": 4529617.376558531,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.078122910087,
+    "estimated_full_shards_per_hour": 65.23596491090044,
     "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": 1.1360651520008105,
-      "bytes": 10573263,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.0926981910015456,
       "bytes": 2481076,
@@ -562,6 +556,12 @@
       "bytes": 6058898,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.2615420420006558,
+      "bytes": 12864855,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -288,3 +288,4 @@
 {"step": 183800, "loss": 0.9128948226571083, "lr": 0.0003, "grad_norm": 0.43129460843954326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26900871117, "tokens_per_second": 147933.23688941126, "data_wait_s": 1.0285999451298267e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 183900, "loss": 0.6273620184510946, "lr": 0.0003, "grad_norm": 0.14109592052623257, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26915408931, "tokens_per_second": 114848.17277801191, "data_wait_s": 1.24769976537209e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 184000, "loss": 0.6542017664760351, "lr": 0.0003, "grad_norm": 0.1794766250336593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26930024755, "tokens_per_second": 148902.9537748219, "data_wait_s": 1.353900006506592e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 184100, "loss": 0.5877811014652252, "lr": 0.0003, "grad_norm": 0.15011607646486622, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26944726361, "tokens_per_second": 150477.16169769233, "data_wait_s": 1.326200072071515e-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 sha256:76b7c1663d13bb2a42638ee62a851a677ebf50f9a83ab0036337bc38d12ed8a6
-size 618300
+oid sha256:0b3650e2e20c9237cc7f6afa43bfe32c8259d35cef117e29339ef9205dfa2e97
+size 620418