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": 75688.01669368563,
-      "required_pages_per_second_upper_estimate": 0.006560445457484151
+      "required_token_positions_per_second": 75998.8852211244,
+      "required_pages_per_second_upper_estimate": 0.006587390753553464
     },
     {
       "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": 9461.002086710703,
-      "required_pages_per_second_upper_estimate": 0.0037741351869757076
+      "required_token_positions_per_second": 9499.86065264055,
+      "required_pages_per_second_upper_estimate": 0.003789636449912458
     },
     {
       "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": 18922.004173421406,
-      "required_pages_per_second_upper_estimate": 0.004377272649621495
+      "required_token_positions_per_second": 18999.7213052811,
+      "required_pages_per_second_upper_estimate": 0.00439525114030243
     },
     {
       "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": 28383.00626013211,
-      "required_pages_per_second_upper_estimate": 0.0017283761671168887
+      "required_token_positions_per_second": 28499.581957921644,
+      "required_pages_per_second_upper_estimate": 0.0017354750154868538
     },
     {
       "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": 28383.00626013211,
-      "required_pages_per_second_upper_estimate": 0.00218500239109495
+      "required_token_positions_per_second": 28499.581957921644,
+      "required_pages_per_second_upper_estimate": 0.0021939767110129736
     },
     {
       "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": 28383.00626013211,
-      "required_pages_per_second_upper_estimate": 0.3524219459395323
+      "required_token_positions_per_second": 28499.581957921644,
+      "required_pages_per_second_upper_estimate": 0.35386942595231563
     }
   ],
   "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": 0.42772661279681756,
-    "producer_p95_seconds": 0.2891327180004737,
-    "gpu_update_seconds": 0.9741040024655694,
-    "producer_batches_per_second": 2.3379419705994042,
-    "gpu_batches_per_second": 1.0265844278114913,
-    "producer_headroom": 2.2773986310931202,
+    "producer_batch_seconds": 0.8173173268517075,
+    "producer_p95_seconds": 0.2937738319997152,
+    "gpu_update_seconds": 0.9701194930094424,
+    "producer_batches_per_second": 1.2235149887889727,
+    "gpu_batches_per_second": 1.0308008520660317,
+    "producer_headroom": 1.1869557406134117,
     "sustainable": true,
-    "buffer_seconds": 60.3944481528653,
+    "buffer_seconds": 60.14740856658543,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 277390.4928969109,
-    "upload_bytes_per_second": 5216832.307734021,
+    "generation_bytes_per_second": 260731.4419515289,
+    "upload_bytes_per_second": 5272328.578830279,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.5215424554872,
+    "estimated_full_shards_per_hour": 55.94689792546654,
     "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": 0.9085387359991728,
-      "bytes": 1084848,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 1.2088771979997546,
       "bytes": 6082441,
@@ -562,6 +556,12 @@
       "bytes": 5972773,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.392480271999375,
+      "bytes": 8010893,
+      "pages": 25,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -194,3 +194,4 @@
 {"step": 256500, "loss": 0.5793410521000624, "lr": 0.0003, "grad_norm": 0.14140056791090877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37544598640, "tokens_per_second": 149039.37765120564, "data_wait_s": 1.3975000911159441e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 256600, "loss": 0.5796477869153023, "lr": 0.0003, "grad_norm": 0.15295621146090357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37559203520, "tokens_per_second": 148569.259743464, "data_wait_s": 1.5387999155791476e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 256700, "loss": 0.5793802235275507, "lr": 0.0003, "grad_norm": 0.1362063822111426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37573826960, "tokens_per_second": 149937.26378596557, "data_wait_s": 1.3355998817132786e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 256800, "loss": 0.5809282548725605, "lr": 0.0003, "grad_norm": 0.16060970723430443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37588422835, "tokens_per_second": 125301.86413422806, "data_wait_s": 1.4684002962894738e-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 sha256:95d8c5d8fa2d2d3595c6b409ac18bc0010987b3e6841aa823e7775757015f0ff
-size 418425
+oid sha256:e44b7c468ff6268feab04195b4a78b5ef9a83673e288331b7e2159ab648470eb
+size 420543