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": 74761.2483165028,
-      "required_pages_per_second_upper_estimate": 0.00648011552342292
+      "required_token_positions_per_second": 73762.97857669221,
+      "required_pages_per_second_upper_estimate": 0.0063935880324676635
     },
     {
       "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": 9345.15603956285,
-      "required_pages_per_second_upper_estimate": 0.0037279224667156733
+      "required_token_positions_per_second": 9220.372322086527,
+      "required_pages_per_second_upper_estimate": 0.003678144376131533
     },
     {
       "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": 18690.3120791257,
-      "required_pages_per_second_upper_estimate": 0.0043236747612477745
+      "required_token_positions_per_second": 18440.744644173054,
+      "required_pages_per_second_upper_estimate": 0.004265941727408322
     },
     {
       "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": 28035.468118688546,
-      "required_pages_per_second_upper_estimate": 0.0017072129176946846
+      "required_token_positions_per_second": 27661.116966259582,
+      "required_pages_per_second_upper_estimate": 0.0016844168965804661
     },
     {
       "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": 28035.468118688546,
-      "required_pages_per_second_upper_estimate": 0.002158247942919475
+      "required_token_positions_per_second": 27661.116966259582,
+      "required_pages_per_second_upper_estimate": 0.0021294293549351793
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28035.468118688546,
-      "required_pages_per_second_upper_estimate": 0.3481066853581403
+      "required_token_positions_per_second": 27661.116966259582,
+      "required_pages_per_second_upper_estimate": 0.3434584969176848
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3290037214764254,
-    "producer_p95_seconds": 0.2800091059980332,
-    "gpu_update_seconds": 0.9861793597649878,
-    "producer_batches_per_second": 3.039479296806843,
-    "gpu_batches_per_second": 1.0140143272095106,
-    "producer_headroom": 2.997471746943908,
+    "producer_batch_seconds": 0.08146025456255757,
+    "producer_p95_seconds": 0.3334741170001507,
+    "gpu_update_seconds": 0.9995257976648021,
+    "producer_batches_per_second": 12.275925300872316,
+    "gpu_batches_per_second": 1.0004744273097361,
+    "producer_headroom": 12.270104028427928,
     "sustainable": true,
-    "buffer_seconds": 60.15694094566426,
+    "buffer_seconds": 60.97107365755293,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 419036.1306905392,
-    "upload_bytes_per_second": 4017915.3324570474,
+    "generation_bytes_per_second": 426153.57915154187,
+    "upload_bytes_per_second": 3969902.3336341484,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.9153989843095,
+    "estimated_full_shards_per_hour": 91.44263773832027,
     "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": 1.0136643189998722,
-      "bytes": 5709641,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.273396311000397,
       "bytes": 6852766,
@@ -562,6 +556,12 @@
       "bytes": 11864781,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.1722596879990306,
+      "bytes": 1208158,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -223,3 +223,4 @@
 {"step": 137300, "loss": 0.5933730937540531, "lr": 0.0003, "grad_norm": 0.1486073915231052, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20093579691, "tokens_per_second": 148801.1000828846, "data_wait_s": 1.5335001080529764e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 137400, "loss": 0.6059115305542946, "lr": 0.0003, "grad_norm": 0.14665909488271753, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20108131900, "tokens_per_second": 148048.114502116, "data_wait_s": 1.317600253969431e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 137500, "loss": 0.5665243081748486, "lr": 0.0003, "grad_norm": 0.13584409338816356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20122669793, "tokens_per_second": 147768.09787086418, "data_wait_s": 1.2518001312855631e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 137600, "loss": 0.5732929743826389, "lr": 0.0003, "grad_norm": 0.1366899343635189, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20137266082, "tokens_per_second": 149694.05483137074, "data_wait_s": 1.8479000573279336e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7bbe161b1bb5e1ea837dff676f332d2c88d3fbc3d7b9f64c8055844f4d8f9584
-size 479847
+oid sha256:38e5f935d4ad2b2fa8c4c3496a8b3e6e58c66bb86e15bc587d4ec6fa4f654019
+size 481965