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": 76048.69766777262,
-      "required_pages_per_second_upper_estimate": 0.00659170837018044
+      "required_token_positions_per_second": 75219.68050786087,
+      "required_pages_per_second_upper_estimate": 0.006519851263884072
     },
     {
       "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": 9506.087208471577,
-      "required_pages_per_second_upper_estimate": 0.0037921203161287607
+      "required_token_positions_per_second": 9402.46006348261,
+      "required_pages_per_second_upper_estimate": 0.0037507818986287733
     },
     {
       "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": 19012.174416943155,
-      "required_pages_per_second_upper_estimate": 0.004398131948518166
+      "required_token_positions_per_second": 18804.92012696522,
+      "required_pages_per_second_upper_estimate": 0.004350187316082696
     },
     {
       "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": 28518.261625414736,
-      "required_pages_per_second_upper_estimate": 0.001736612509232542
+      "required_token_positions_per_second": 28207.380190447828,
+      "required_pages_per_second_upper_estimate": 0.0017176814609118909
     },
     {
       "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": 28518.261625414736,
-      "required_pages_per_second_upper_estimate": 0.0021954147235252208
+      "required_token_positions_per_second": 28207.380190447828,
+      "required_pages_per_second_upper_estimate": 0.002171482210086577
     },
     {
       "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": 28518.261625414736,
-      "required_pages_per_second_upper_estimate": 0.35410136490575433
+      "required_token_positions_per_second": 28207.380190447828,
+      "required_pages_per_second_upper_estimate": 0.35024125793669775
     }
   ],
   "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": 0.2925638286401977,
-    "producer_p95_seconds": 0.3399124580028001,
-    "gpu_update_seconds": 0.9694840577295504,
-    "producer_batches_per_second": 3.418057538581863,
-    "gpu_batches_per_second": 1.0314764766136695,
-    "producer_headroom": 3.3137522920574236,
+    "producer_batch_seconds": 0.10926558689777721,
+    "producer_p95_seconds": 0.33388570899842307,
+    "gpu_update_seconds": 0.9801690129791899,
+    "producer_batches_per_second": 9.152012343424689,
+    "gpu_batches_per_second": 1.020232211749415,
+    "producer_headroom": 8.97051890542794,
     "sustainable": true,
-    "buffer_seconds": 60.10801157923213,
+    "buffer_seconds": 60.770478804709775,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 346442.4156242738,
-    "upload_bytes_per_second": 4510056.934459141,
+    "generation_bytes_per_second": 345444.16180377884,
+    "upload_bytes_per_second": 4466181.861380511,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.33847762628709,
+    "estimated_full_shards_per_hour": 74.12427559456849,
     "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": 1.2234259870019741,
-      "bytes": 10476491,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.209228235999035,
       "bytes": 11665620,
@@ -562,6 +556,12 @@
       "bytes": 12956908,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 1.2346246180022717,
+      "bytes": 6315095,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -342,3 +342,4 @@
 {"step": 189200, "loss": 0.5980932433158159, "lr": 0.0003, "grad_norm": 0.1434137600885681, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27690592215, "tokens_per_second": 142521.74885826552, "data_wait_s": 8.87100031832233e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 189300, "loss": 0.5943525526672602, "lr": 0.0003, "grad_norm": 0.15447451011079089, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27705116481, "tokens_per_second": 148907.89057337184, "data_wait_s": 1.2247997801750898e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 189400, "loss": 0.5764011181890965, "lr": 0.0003, "grad_norm": 0.1454088546957294, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27719889100, "tokens_per_second": 151355.7005699601, "data_wait_s": 1.0283998562954366e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 189500, "loss": 0.5868939906358719, "lr": 0.0003, "grad_norm": 0.1563980391468381, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27734569853, "tokens_per_second": 151238.77286017945, "data_wait_s": 1.2299999070819467e-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 sha256:0cccff0c17ee64b88658ef6acd29575d1d03fcf4b80dfc887d0dcb7646aa6b30
-size 733455
+oid sha256:ee83a9be4d5ffab0a9d8084a2fa31a4f4cdaaea2f0dbffc6c9508075a3172674
+size 735573