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": 75094.7918297736,
-      "required_pages_per_second_upper_estimate": 0.006509026229794928
+      "required_token_positions_per_second": 75781.77679983915,
+      "required_pages_per_second_upper_estimate": 0.006568572345852721
     },
     {
       "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": 9386.8489787217,
-      "required_pages_per_second_upper_estimate": 0.00374455440351113
+      "required_token_positions_per_second": 9472.722099979894,
+      "required_pages_per_second_upper_estimate": 0.0037788104754986015
     },
     {
       "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": 18773.6979574434,
-      "required_pages_per_second_upper_estimate": 0.004342964616655239
+      "required_token_positions_per_second": 18945.444199959788,
+      "required_pages_per_second_upper_estimate": 0.004382695087230773
     },
     {
       "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": 28160.5469361651,
-      "required_pages_per_second_upper_estimate": 0.0017148295614411683
+      "required_token_positions_per_second": 28418.16629993968,
+      "required_pages_per_second_upper_estimate": 0.0017305172290706973
     },
     {
       "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": 28160.5469361651,
-      "required_pages_per_second_upper_estimate": 0.0021678768565291475
+      "required_token_positions_per_second": 28418.16629993968,
+      "required_pages_per_second_upper_estimate": 0.002187709108288556
     },
     {
       "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": 28160.5469361651,
-      "required_pages_per_second_upper_estimate": 0.34965974565932556
+      "required_token_positions_per_second": 28418.16629993968,
+      "required_pages_per_second_upper_estimate": 0.3528585159608587
     }
   ],
   "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": 0.0786415333516004,
-    "producer_p95_seconds": 0.19592558900149015,
-    "gpu_update_seconds": 0.9817991128749398,
-    "producer_batches_per_second": 12.715927034752426,
-    "gpu_batches_per_second": 1.0185383006425457,
-    "producer_headroom": 12.484485882102394,
+    "producer_batch_seconds": 0.09003981251559878,
+    "producer_p95_seconds": 0.1951486029993248,
+    "gpu_update_seconds": 0.9728988038210856,
+    "producer_batches_per_second": 11.10619815902834,
+    "gpu_batches_per_second": 1.0278561306401794,
+    "producer_headroom": 10.805206903918615,
     "sustainable": true,
-    "buffer_seconds": 60.87154499824627,
+    "buffer_seconds": 60.31972583690731,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 435772.6872696558,
-    "upload_bytes_per_second": 5278880.475192897,
+    "generation_bytes_per_second": 415550.65695912385,
+    "upload_bytes_per_second": 5184286.961140318,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.5066744190908,
+    "estimated_full_shards_per_hour": 89.16749745922363,
     "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": 1.3712029609996534,
-      "bytes": 12398330,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 0.7755766290001702,
       "bytes": 1144894,
@@ -562,6 +556,12 @@
       "bytes": 5968582,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.9677518210010021,
+      "bytes": 2711632,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -119,3 +119,4 @@
 {"step": 166900, "loss": 0.6116734612733126, "lr": 0.0003, "grad_norm": 0.14229868911171528, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24426748624, "tokens_per_second": 150449.8063104413, "data_wait_s": 1.085700023395475e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 167000, "loss": 0.5920278616249561, "lr": 0.0003, "grad_norm": 0.13092271103070144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24441291686, "tokens_per_second": 148720.9678397098, "data_wait_s": 1.6636000509606674e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 167100, "loss": 0.641258591786027, "lr": 0.0003, "grad_norm": 0.16054336133288577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24455944115, "tokens_per_second": 150415.35592970694, "data_wait_s": 1.2121001418563537e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 167200, "loss": 0.5853394120931625, "lr": 0.0003, "grad_norm": 0.14848904564804077, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24470597460, "tokens_per_second": 149105.21808988327, "data_wait_s": 8.761999197304249e-06, "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:c7ab82a84eef1bb48e61dbdfbc3afc3ed1158132bbfcc73319912711c8fcc0a7
-size 258009
+oid sha256:2a31ad0a423ef83b6bfcbed04aeeb60ed1fb04b77953bd303ed8b27d04465f05
+size 260127