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": 75766.5250573859,
-      "required_pages_per_second_upper_estimate": 0.006567250363472062
+      "required_token_positions_per_second": 75701.50907361948,
+      "required_pages_per_second_upper_estimate": 0.006561614942780693
     },
     {
       "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": 9470.815632173237,
-      "required_pages_per_second_upper_estimate": 0.0037780499569862918
+      "required_token_positions_per_second": 9462.688634202435,
+      "required_pages_per_second_upper_estimate": 0.0037748079759862916
     },
     {
       "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": 18941.631264346473,
-      "required_pages_per_second_upper_estimate": 0.004381813031682007
+      "required_token_positions_per_second": 18925.37726840487,
+      "required_pages_per_second_upper_estimate": 0.004378052955781476
     },
     {
       "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": 28412.44689651971,
-      "required_pages_per_second_upper_estimate": 0.0017301689474098117
+      "required_token_positions_per_second": 28388.065902607304,
+      "required_pages_per_second_upper_estimate": 0.0017286842727977354
     },
     {
       "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": 28412.44689651971,
-      "required_pages_per_second_upper_estimate": 0.0021872688127809645
+      "required_token_positions_per_second": 28388.065902607304,
+      "required_pages_per_second_upper_estimate": 0.002185391896378677
     },
     {
       "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": 28412.44689651971,
-      "required_pages_per_second_upper_estimate": 0.3527875001119946
+      "required_token_positions_per_second": 28388.065902607304,
+      "required_pages_per_second_upper_estimate": 0.3524847697655401
     }
   ],
   "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": 0.42342813355463704,
-    "producer_p95_seconds": 0.33924947700143093,
-    "gpu_update_seconds": 0.9730946475921668,
-    "producer_batches_per_second": 2.361675856549964,
-    "gpu_batches_per_second": 1.0276492656437974,
-    "producer_headroom": 2.298134135356416,
+    "producer_batch_seconds": 0.09025718112499703,
+    "producer_p95_seconds": 0.28312330299741006,
+    "gpu_update_seconds": 0.973930386622805,
+    "producer_batches_per_second": 11.07945082635698,
+    "gpu_batches_per_second": 1.0267674299264795,
+    "producer_headroom": 10.79061382688221,
     "sustainable": true,
-    "buffer_seconds": 60.33186815071434,
+    "buffer_seconds": 60.38368397061391,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297320.828917994,
-    "upload_bytes_per_second": 5595141.8590854,
+    "generation_bytes_per_second": 313596.99434861523,
+    "upload_bytes_per_second": 5653046.756379689,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.798128611134196,
+    "estimated_full_shards_per_hour": 67.29061482280581,
     "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": 0.8625212450006075,
-      "bytes": 879863,
-      "pages": 25,
-      "shards": 1
-    },
     {
       "seconds": 1.04706289600108,
       "bytes": 6208354,
@@ -562,6 +556,12 @@
       "bytes": 6612201,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.0170475559971237,
+      "bytes": 6061997,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -220,3 +220,4 @@
 {"step": 377000, "loss": 0.5250621605664492, "lr": 0.0003, "grad_norm": 0.1611399413595201, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55182000710, "tokens_per_second": 149231.57059292105, "data_wait_s": 1.909200000227429e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 377100, "loss": 0.5500581096857786, "lr": 0.0003, "grad_norm": 0.14925368890034985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55196456799, "tokens_per_second": 141692.68027099827, "data_wait_s": 1.0667001333786175e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 377200, "loss": 0.5740863550454378, "lr": 0.0003, "grad_norm": 0.16492200842018148, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55211052911, "tokens_per_second": 148272.81830550425, "data_wait_s": 1.8700000509852543e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 377300, "loss": 0.5944837853312492, "lr": 0.0003, "grad_norm": 0.1480909025327743, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55225745724, "tokens_per_second": 149216.04525386522, "data_wait_s": 1.3347998901735991e-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 sha256:37ad851207f1cf0641d3dc206b8368ee1bd175903f91a886fcad9b66c31b46e5
-size 473493
+oid sha256:c1619c8bfb1870008fe4e7ae743f364e5424798e4098195b0394cc74a3427f25
+size 475611