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": 76096.92891025094,
-      "required_pages_per_second_upper_estimate": 0.0065958889320374935
+      "required_token_positions_per_second": 75674.70397068192,
+      "required_pages_per_second_upper_estimate": 0.006559291544394977
     },
     {
       "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": 9512.116113781367,
-      "required_pages_per_second_upper_estimate": 0.003794525336597003
+      "required_token_positions_per_second": 9459.33799633524,
+      "required_pages_per_second_upper_estimate": 0.0037734713564445666
     },
     {
       "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": 19024.232227562734,
-      "required_pages_per_second_upper_estimate": 0.004400921310794787
+      "required_token_positions_per_second": 18918.67599267048,
+      "required_pages_per_second_upper_estimate": 0.004376502733578749
     },
     {
       "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": 28536.3483413441,
-      "required_pages_per_second_upper_estimate": 0.0017377138953389754
+      "required_token_positions_per_second": 28378.01398900572,
+      "required_pages_per_second_upper_estimate": 0.0017280721639977138
     },
     {
       "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": 28536.3483413441,
-      "required_pages_per_second_upper_estimate": 0.0021968070889846975
+      "required_token_positions_per_second": 28378.01398900572,
+      "required_pages_per_second_upper_estimate": 0.0021846180722441496
     },
     {
       "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": 28536.3483413441,
-      "required_pages_per_second_upper_estimate": 0.35432594138525275
+      "required_token_positions_per_second": 28378.01398900572,
+      "required_pages_per_second_upper_estimate": 0.3523599586401992
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.2238003082110822,
-    "producer_p95_seconds": 0.3680593989993213,
-    "gpu_update_seconds": 0.9688695858798079,
-    "producer_batches_per_second": 4.468269092180284,
-    "gpu_batches_per_second": 1.0321306547071796,
-    "producer_headroom": 4.3291700249402565,
+    "producer_batch_seconds": 0.1398951794217851,
+    "producer_p95_seconds": 0.37108688300213544,
+    "gpu_update_seconds": 0.9742753672160235,
+    "producer_batches_per_second": 7.148209138679411,
+    "gpu_batches_per_second": 1.0264038624495704,
+    "producer_headroom": 6.964324083523818,
     "sustainable": true,
-    "buffer_seconds": 60.06991432454809,
+    "buffer_seconds": 60.405072767393456,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 373735.20479932614,
-    "upload_bytes_per_second": 5202283.592307462,
+    "generation_bytes_per_second": 374566.7437154129,
+    "upload_bytes_per_second": 5276435.037031733,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.19487483963812,
+    "estimated_full_shards_per_hour": 80.37330373379507,
     "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": 0.8931559880002169,
-      "bytes": 949478,
-      "pages": 28,
-      "shards": 1
-    },
     {
       "seconds": 1.1832785620026698,
       "bytes": 12385340,
@@ -562,6 +556,12 @@
       "bytes": 11430289,
       "pages": 42,
       "shards": 3
+    },
+    {
+      "seconds": 1.0073759259976214,
+      "bytes": 7448462,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -369,3 +369,4 @@
 {"step": 391900, "loss": 0.5476494245231152, "lr": 0.0003, "grad_norm": 0.14311585357797804, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57363209465, "tokens_per_second": 131533.90772729163, "data_wait_s": 9.354000212624669e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 392000, "loss": 0.5833262894302607, "lr": 0.0003, "grad_norm": 0.1442648034579037, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57377873761, "tokens_per_second": 149932.2662058907, "data_wait_s": 1.5614998119417578e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 392100, "loss": 0.5720165120437741, "lr": 0.0003, "grad_norm": 0.14173155540314036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57392644680, "tokens_per_second": 150432.58370370566, "data_wait_s": 9.858995326794684e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 392200, "loss": 0.5652673803269863, "lr": 0.0003, "grad_norm": 0.17720953343374488, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57407343904, "tokens_per_second": 149683.16081492504, "data_wait_s": 1.5159996110014617e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bb52d258f94eb7cff954bad68156a059032d6d893c92e5f1c3eddbe7cf88e95c
-size 791424
+oid sha256:3466d053dd423f5290fee33a9a263401bed8d50fcb4a3ffd965f11c8b5d51601
+size 793542