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": 75650.14668849629,
-      "required_pages_per_second_upper_estimate": 0.006557162981414989
+      "required_token_positions_per_second": 75312.28636741362,
+      "required_pages_per_second_upper_estimate": 0.006527878105082699
     },
     {
       "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": 9456.268336062036,
-      "required_pages_per_second_upper_estimate": 0.0037722468230660744
+      "required_token_positions_per_second": 9414.035795926702,
+      "required_pages_per_second_upper_estimate": 0.0037553996313733455
     },
     {
       "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": 18912.53667212407,
-      "required_pages_per_second_upper_estimate": 0.004375082509818689
+      "required_token_positions_per_second": 18828.071591853404,
+      "required_pages_per_second_upper_estimate": 0.004355543000032712
     },
     {
       "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": 28368.805008186107,
-      "required_pages_per_second_upper_estimate": 0.0017275113853815877
+      "required_token_positions_per_second": 28242.10738778011,
+      "required_pages_per_second_upper_estimate": 0.0017197961650299036
     },
     {
       "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": 28368.805008186107,
-      "required_pages_per_second_upper_estimate": 0.002183909139408565
+      "required_token_positions_per_second": 28242.10738778011,
+      "required_pages_per_second_upper_estimate": 0.002174155605868251
     },
     {
       "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": 28368.805008186107,
-      "required_pages_per_second_upper_estimate": 0.35224561391889575
+      "required_token_positions_per_second": 28242.10738778011,
+      "required_pages_per_second_upper_estimate": 0.3506724535031117
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.12850909054657222,
-    "producer_p95_seconds": 0.3594871709938161,
-    "gpu_update_seconds": 0.9745916330286697,
-    "producer_batches_per_second": 7.781550672772024,
-    "gpu_batches_per_second": 1.0260707829928424,
-    "producer_headroom": 7.58383417767223,
+    "producer_batch_seconds": 0.4339878128900523,
+    "producer_p95_seconds": 0.37877907900110586,
+    "gpu_update_seconds": 0.9789637727941942,
+    "producer_batches_per_second": 2.3042121697858433,
+    "gpu_batches_per_second": 1.02148825910663,
+    "producer_headroom": 2.2557402390518457,
     "sustainable": true,
-    "buffer_seconds": 60.42468124777752,
+    "buffer_seconds": 60.69575391324004,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 303431.3106896626,
-    "upload_bytes_per_second": 5249503.450534328,
+    "generation_bytes_per_second": 281622.28446901665,
+    "upload_bytes_per_second": 5232234.624428951,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.10929575459869,
+    "estimated_full_shards_per_hour": 60.429586415795086,
     "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": 1.1763414359993476,
-      "bytes": 9929012,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.3798298289984814,
       "bytes": 7738436,
@@ -562,6 +556,12 @@
       "bytes": 2697194,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.4709027479984798,
+      "bytes": 10099926,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -369,3 +369,4 @@
 {"step": 274000, "loss": 0.5780625939369202, "lr": 0.0003, "grad_norm": 0.14163760091911487, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40105112994, "tokens_per_second": 149696.021686254, "data_wait_s": 1.6744001186452806e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 274100, "loss": 0.5045449025928974, "lr": 0.0003, "grad_norm": 0.15994933381579993, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40120017579, "tokens_per_second": 152369.77918447362, "data_wait_s": 1.2128999514970928e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 274200, "loss": 0.6100579909980297, "lr": 0.0003, "grad_norm": 0.1579126367697439, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40134537871, "tokens_per_second": 147906.9049218179, "data_wait_s": 1.1185002222191542e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 274300, "loss": 0.5829500239342451, "lr": 0.0003, "grad_norm": 0.12775942397489143, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40148945799, "tokens_per_second": 147073.15038042015, "data_wait_s": 1.310199877480045e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9297d3ffacb400d0674ad9754e8f3a239984a4490ec525d7ce285e2d92f0d210
-size 791424
+oid sha256:d80c4002e763b553df31f3a2be72bb574366f8c64cc16710eb4a1e144140dcbb
+size 793542