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": 75209.1618081414,
-      "required_pages_per_second_upper_estimate": 0.006518939529651789
+      "required_token_positions_per_second": 73782.36947322769,
+      "required_pages_per_second_upper_estimate": 0.006395268786233581
     },
     {
       "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": 9401.145226017676,
-      "required_pages_per_second_upper_estimate": 0.0037502573903054393
+      "required_token_positions_per_second": 9222.796184153462,
+      "required_pages_per_second_upper_estimate": 0.0036791112909500007
     },
     {
       "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": 18802.29045203535,
-      "required_pages_per_second_upper_estimate": 0.004349578987068369
+      "required_token_positions_per_second": 18445.592368306923,
+      "required_pages_per_second_upper_estimate": 0.004267063163069662
     },
     {
       "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": 28203.43567805303,
-      "required_pages_per_second_upper_estimate": 0.0017174412608023048
+      "required_token_positions_per_second": 27668.388552460387,
+      "required_pages_per_second_upper_estimate": 0.001684859698029031
     },
     {
       "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": 28203.43567805303,
-      "required_pages_per_second_upper_estimate": 0.002171178550603316
+      "required_token_positions_per_second": 27668.388552460387,
+      "required_pages_per_second_upper_estimate": 0.002129989141769956
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28203.43567805303,
-      "required_pages_per_second_upper_estimate": 0.35019228029418814
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27668.388552460387,
+      "required_pages_per_second_upper_estimate": 0.3435487856818653
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13910729061740312,
-    "producer_p95_seconds": 0.36179056499895523,
-    "gpu_update_seconds": 0.9803060987181342,
-    "producer_batches_per_second": 7.188695830115566,
-    "gpu_batches_per_second": 1.020089542753654,
-    "producer_headroom": 7.047122364091909,
+    "producer_batch_seconds": 0.0871788646484788,
+    "producer_p95_seconds": 0.353184349998628,
+    "gpu_update_seconds": 0.9992631102305893,
+    "producer_batches_per_second": 11.470670145019481,
+    "gpu_batches_per_second": 1.0007374331763739,
+    "producer_headroom": 11.462217525541332,
     "sustainable": true,
-    "buffer_seconds": 60.77897812052432,
+    "buffer_seconds": 60.95504972406595,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 91,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 323306.75751933106,
-    "upload_bytes_per_second": 4454698.648017714,
+    "generation_bytes_per_second": 336295.6182658304,
+    "upload_bytes_per_second": 4497382.874200761,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.37410396752308,
+    "estimated_full_shards_per_hour": 72.16121111851868,
     "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": 0.9047177330030536,
-      "bytes": 2631324,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.3315005119984562,
       "bytes": 5609510,
@@ -562,6 +556,12 @@
       "bytes": 957902,
       "pages": 28,
       "shards": 1
+    },
+    {
+      "seconds": 1.1445572449993051,
+      "bytes": 7733205,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -286,3 +286,4 @@
 {"step": 183600, "loss": 0.614504911005497, "lr": 0.0003, "grad_norm": 0.14576641525086484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26871718235, "tokens_per_second": 150524.31812238303, "data_wait_s": 2.158799907192588e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 183700, "loss": 0.6800808813422918, "lr": 0.0003, "grad_norm": 0.16859660292366452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26886354630, "tokens_per_second": 149794.669351485, "data_wait_s": 7.031998393358663e-06, "gpu_peak_gb": 85.46427917480469}
 {"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}

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 sha256:476ca0712e2c7b18760eed95bc6d35de079578f65b8b4685a693bec595e2b486
-size 614064
+oid sha256:c712c0e13a4d0b7d564915ae2e19f7306eb3ac8736d0531aa5cf7fe2ff56803d
+size 616182