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": 74770.47765798795,
-      "required_pages_per_second_upper_estimate": 0.006480915499351302
+      "required_token_positions_per_second": 75499.42071071903,
+      "required_pages_per_second_upper_estimate": 0.006544098435672755
     },
     {
       "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": 9346.309707248494,
-      "required_pages_per_second_upper_estimate": 0.003728382682004346
+      "required_token_positions_per_second": 9437.427588839879,
+      "required_pages_per_second_upper_estimate": 0.0037647309673048824
     },
     {
       "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": 18692.61941449699,
-      "required_pages_per_second_upper_estimate": 0.004324208522678207
+      "required_token_positions_per_second": 18874.855177679758,
+      "required_pages_per_second_upper_estimate": 0.004366365559250649
     },
     {
       "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": 28038.929121745485,
-      "required_pages_per_second_upper_estimate": 0.0017074236746223725
+      "required_token_positions_per_second": 28312.282766519635,
+      "required_pages_per_second_upper_estimate": 0.001724069477413383
     },
     {
       "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": 28038.929121745485,
-      "required_pages_per_second_upper_estimate": 0.002158514380515468
+      "required_token_positions_per_second": 28312.282766519635,
+      "required_pages_per_second_upper_estimate": 0.002179557900781501
     },
     {
       "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": 28038.929121745485,
-      "required_pages_per_second_upper_estimate": 0.3481496594328754
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28312.282766519635,
+      "required_pages_per_second_upper_estimate": 0.35154379684517223
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.05695632236715653,
-    "producer_p95_seconds": 0.1906214450009429,
-    "gpu_update_seconds": 0.9860576300882226,
-    "producer_batches_per_second": 17.55731336643749,
-    "gpu_batches_per_second": 1.0141395081649842,
-    "producer_headroom": 17.312522808825626,
+    "producer_batch_seconds": 0.3980550660389639,
+    "producer_p95_seconds": 0.2091063239986397,
+    "gpu_update_seconds": 0.9765372940077733,
+    "producer_batches_per_second": 2.5122152318044213,
+    "gpu_batches_per_second": 1.024026431080716,
+    "producer_headroom": 2.4532718644314007,
     "sustainable": true,
-    "buffer_seconds": 60.14951543538157,
+    "buffer_seconds": 60.54531222848195,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342675.44495674816,
-    "upload_bytes_per_second": 5844408.80102132,
+    "generation_bytes_per_second": 322619.42491805827,
+    "upload_bytes_per_second": 5719672.202854442,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.53017341162523,
+    "estimated_full_shards_per_hour": 69.22661839157401,
     "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": [
@@ -286,6 +286,12 @@
       "bytes": 7647354,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.2463584760007507,
+      "bytes": 4721254,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -68,3 +68,4 @@
 {"step": 91800, "loss": 0.5888327732682228, "lr": 0.0003, "grad_norm": 0.12957206124342208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13434264611, "tokens_per_second": 150189.06462793087, "data_wait_s": 1.0155999916605651e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 91900, "loss": 0.5598646011203527, "lr": 0.0003, "grad_norm": 0.1423399720042594, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13449024303, "tokens_per_second": 150284.03344551637, "data_wait_s": 1.3726999895879999e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 92000, "loss": 0.6390262767672539, "lr": 0.0003, "grad_norm": 0.1441192939934209, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13463728262, "tokens_per_second": 149921.50896801552, "data_wait_s": 9.022000085678883e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 92100, "loss": 0.6326837167143822, "lr": 0.0003, "grad_norm": 0.15781426775344015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13478260656, "tokens_per_second": 147960.2941388051, "data_wait_s": 1.72180007211864e-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:b302d39b6f302b7d81b1e00f9d0069889461bcd015ae8830bc027787ba1022ec
-size 149208
+oid sha256:68e5d05ec61111e73ea6c64c8423f1ebab7494df02f4a357fe2d62d42f631774
+size 151326