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": 72881.86046424118,
-      "required_pages_per_second_upper_estimate": 0.006317214947653848
+      "required_token_positions_per_second": 75750.67792838323,
+      "required_pages_per_second_upper_estimate": 0.006565876774494292
     },
     {
       "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": 9110.232558030148,
-      "required_pages_per_second_upper_estimate": 0.003634207977513223
+      "required_token_positions_per_second": 9468.834741047904,
+      "required_pages_per_second_upper_estimate": 0.003777259749899435
     },
     {
       "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": 18220.465116060295,
-      "required_pages_per_second_upper_estimate": 0.004214983935366721
+      "required_token_positions_per_second": 18937.669482095807,
+      "required_pages_per_second_upper_estimate": 0.004380896543083302
     },
     {
       "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": 27330.697674090447,
-      "required_pages_per_second_upper_estimate": 0.0016642960952634141
+      "required_token_positions_per_second": 28406.50422314371,
+      "required_pages_per_second_upper_estimate": 0.001729807069780011
     },
     {
       "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": 27330.697674090447,
-      "required_pages_per_second_upper_estimate": 0.0021039927631648575
+      "required_token_positions_per_second": 28406.50422314371,
+      "required_pages_per_second_upper_estimate": 0.0021868113293341076
     },
     {
       "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": 27330.697674090447,
-      "required_pages_per_second_upper_estimate": 0.3393557951511783
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28406.50422314371,
+      "required_pages_per_second_upper_estimate": 0.35271371199751306
     }
   ],
   "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": 60,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.44935046455483985,
-    "producer_p95_seconds": 0.3586128569986613,
-    "gpu_update_seconds": 1.011609741166994,
-    "producer_batches_per_second": 2.2254344412232325,
-    "gpu_batches_per_second": 0.9885234980501463,
-    "producer_headroom": 2.2512711590699483,
+    "producer_batch_seconds": 0.08887273362506676,
+    "producer_p95_seconds": 0.34691206400020747,
+    "gpu_update_seconds": 0.9732982201123596,
+    "producer_batches_per_second": 11.252045022253572,
+    "gpu_batches_per_second": 1.0274343252005105,
+    "producer_headroom": 10.951595392783537,
     "sustainable": true,
-    "buffer_seconds": 60.69658447001964,
+    "buffer_seconds": 60.344489646966295,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 431258.0870426268,
-    "upload_bytes_per_second": 4574155.61749292,
+    "generation_bytes_per_second": 430865.95374321163,
+    "upload_bytes_per_second": 4543113.590942459,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.53794630488494,
+    "estimated_full_shards_per_hour": 92.45380362722646,
     "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": 1.1226200919991243,
-      "bytes": 7462663,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.8068668020023324,
       "bytes": 982732,
@@ -562,6 +556,12 @@
       "bytes": 10104452,
       "pages": 27,
       "shards": 3
+    },
+    {
+      "seconds": 1.954254078998929,
+      "bytes": 8371700,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -318,3 +318,4 @@
 {"step": 186800, "loss": 0.633107615634799, "lr": 0.0003, "grad_norm": 0.14072124692690208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27339416525, "tokens_per_second": 148882.1327731205, "data_wait_s": 1.079800131265074e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 186900, "loss": 0.587534736841917, "lr": 0.0003, "grad_norm": 0.14738928761525955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27353807895, "tokens_per_second": 145889.4209428797, "data_wait_s": 1.0378000297350809e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 187000, "loss": 0.5876833312213421, "lr": 0.0003, "grad_norm": 0.13868859085156893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27368515947, "tokens_per_second": 151428.8942013001, "data_wait_s": 1.1518000974319875e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 187100, "loss": 0.6052984334528446, "lr": 0.0003, "grad_norm": 0.12999672994602687, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27383160466, "tokens_per_second": 149658.03758488997, "data_wait_s": 1.3742999726673588e-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 sha256:9a9db4117f92f939f7c43e6738df813e49bc0f64bbd82308388aa5c7111db5da
-size 682623
+oid sha256:76cf25fec2c64f1256833094c244de790aeb306640463e05dce47fc0bd3002f2
+size 684741