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": 75156.41678357421,
-      "required_pages_per_second_upper_estimate": 0.006514367724603348
+      "required_token_positions_per_second": 75530.25830100315,
+      "required_pages_per_second_upper_estimate": 0.006546771359841425
     },
     {
       "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": 9394.552097946776,
-      "required_pages_per_second_upper_estimate": 0.0037476272929419085
+      "required_token_positions_per_second": 9441.282287625394,
+      "required_pages_per_second_upper_estimate": 0.003766268664283307
     },
     {
       "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": 18789.104195893553,
-      "required_pages_per_second_upper_estimate": 0.004346528578780146
+      "required_token_positions_per_second": 18882.564575250788,
+      "required_pages_per_second_upper_estimate": 0.0043681489926979435
     },
     {
       "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": 28183.656293840326,
-      "required_pages_per_second_upper_estimate": 0.00171623680007816
+      "required_token_positions_per_second": 28323.846862876184,
+      "required_pages_per_second_upper_estimate": 0.0017247736702093721
     },
     {
       "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": 28183.656293840326,
-      "required_pages_per_second_upper_estimate": 0.0021696558788537827
+      "required_token_positions_per_second": 28323.846862876184,
+      "required_pages_per_second_upper_estimate": 0.0021804481369304958
     },
     {
       "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": 28183.656293840326,
-      "required_pages_per_second_upper_estimate": 0.349946686539607
+      "required_token_positions_per_second": 28323.846862876184,
+      "required_pages_per_second_upper_estimate": 0.35168738421937973
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3650547258515502,
-    "producer_p95_seconds": 0.21641876899957424,
-    "gpu_update_seconds": 0.9809940808156464,
-    "producer_batches_per_second": 2.739315311334035,
-    "gpu_batches_per_second": 1.0193741425723497,
-    "producer_headroom": 2.687252105906358,
+    "producer_batch_seconds": 0.08423081715631042,
+    "producer_p95_seconds": 0.20517177799956698,
+    "gpu_update_seconds": 0.9761385921146887,
+    "producer_batches_per_second": 11.872139363723148,
+    "gpu_batches_per_second": 1.024444692667686,
+    "producer_headroom": 11.588853403894088,
     "sustainable": true,
-    "buffer_seconds": 60.82163301057008,
+    "buffer_seconds": 60.520592711110694,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 306712.2469209871,
-    "upload_bytes_per_second": 4952911.046242599,
+    "generation_bytes_per_second": 307915.45102200005,
+    "upload_bytes_per_second": 5642331.908642662,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.8133082935544,
+    "estimated_full_shards_per_hour": 66.07148788447381,
     "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": [
@@ -232,6 +232,12 @@
       "bytes": 1081901,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.247454252999887,
+      "bytes": 13520201,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -57,3 +57,4 @@
 {"step": 90700, "loss": 0.5889896973967552, "lr": 0.0003, "grad_norm": 0.13338667831945086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13273557530, "tokens_per_second": 149663.66695066163, "data_wait_s": 1.3901000784244388e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 90800, "loss": 0.6065632402896881, "lr": 0.0003, "grad_norm": 0.1261719028856433, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13288072735, "tokens_per_second": 147650.11146710603, "data_wait_s": 9.131999831879511e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 90900, "loss": 0.6078984197229147, "lr": 0.0003, "grad_norm": 0.13313520223083922, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13302686830, "tokens_per_second": 149111.6854930715, "data_wait_s": 1.4524999642162584e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 91000, "loss": 0.6044453363865614, "lr": 0.0003, "grad_norm": 0.13011336553444902, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13317251419, "tokens_per_second": 147588.44052954204, "data_wait_s": 1.0572000974207185e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5e956d66b09e352a7675d6e1c234b2adec23407d30c972c3920b9cfcffd24f7b
-size 125910
+oid sha256:bbaad4bb0706e6dd7e605b70c598cfb96ab48a896bcffd7d366e898c0aa143a8
+size 128028