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": 74826.87545340792,
-      "required_pages_per_second_upper_estimate": 0.0064858039173193984
+      "required_token_positions_per_second": 74906.60831207344,
+      "required_pages_per_second_upper_estimate": 0.006492714959427443
     },
     {
       "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": 9353.35943167599,
-      "required_pages_per_second_upper_estimate": 0.003731194922481247
+      "required_token_positions_per_second": 9363.32603900918,
+      "required_pages_per_second_upper_estimate": 0.003735170751160515
     },
     {
       "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": 18706.71886335198,
-      "required_pages_per_second_upper_estimate": 0.004327470182028984
+      "required_token_positions_per_second": 18726.65207801836,
+      "required_pages_per_second_upper_estimate": 0.004332081380429455
     },
     {
       "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": 28060.078295027968,
-      "required_pages_per_second_upper_estimate": 0.001708711548314139
+      "required_token_positions_per_second": 28089.978117027542,
+      "required_pages_per_second_upper_estimate": 0.0017105322906016708
     },
     {
       "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": 28060.078295027968,
-      "required_pages_per_second_upper_estimate": 0.002160142502419412
+      "required_token_positions_per_second": 28089.978117027542,
+      "required_pages_per_second_upper_estimate": 0.002162444273484944
     },
     {
       "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": 28060.078295027968,
-      "required_pages_per_second_upper_estimate": 0.3484122613833141
+      "required_token_positions_per_second": 28089.978117027542,
+      "required_pages_per_second_upper_estimate": 0.3487835171042818
     }
   ],
   "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": 0.08451387303122715,
-    "producer_p95_seconds": 0.2664653850006289,
-    "gpu_update_seconds": 0.985314428181728,
-    "producer_batches_per_second": 11.832376912019031,
-    "gpu_batches_per_second": 1.0149044522217872,
-    "producer_headroom": 11.658611691096713,
+    "producer_batch_seconds": 0.4761744179136542,
+    "producer_p95_seconds": 0.29677545099912095,
+    "gpu_update_seconds": 0.9842656297136941,
+    "producer_batches_per_second": 2.1000708193889834,
+    "gpu_batches_per_second": 1.0159858983299892,
+    "producer_headroom": 2.0670275274892513,
     "sustainable": true,
-    "buffer_seconds": 60.104180119085406,
+    "buffer_seconds": 60.04020341253534,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 290587.8447057929,
-    "upload_bytes_per_second": 5266415.1016032,
+    "generation_bytes_per_second": 270360.7942323195,
+    "upload_bytes_per_second": 5323039.205009043,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.35338693504658,
+    "estimated_full_shards_per_hour": 58.01313276507558,
     "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": 0.723643687000731,
-      "bytes": 1209843,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.586552576000031,
       "bytes": 9572867,
@@ -562,6 +556,12 @@
       "bytes": 11010766,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.1728442359999462,
+      "bytes": 8097014,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -180,3 +180,4 @@
 {"step": 213000, "loss": 0.5666066184639931, "lr": 0.0003, "grad_norm": 0.1961237770430454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31176111509, "tokens_per_second": 146116.2680647569, "data_wait_s": 1.881700154626742e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 213100, "loss": 0.6173598244786263, "lr": 0.0003, "grad_norm": 0.14808592038727653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31190635549, "tokens_per_second": 147841.02489907615, "data_wait_s": 1.7596001271158457e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 213200, "loss": 0.5844620242714882, "lr": 0.0003, "grad_norm": 0.1403808011301817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31205436021, "tokens_per_second": 129244.27535796976, "data_wait_s": 2.584400135674514e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 213300, "loss": 0.5944621972739697, "lr": 0.0003, "grad_norm": 0.16282966041403288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31220207452, "tokens_per_second": 149827.3292518399, "data_wait_s": 1.5639998309779912e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f7cf3712970ef12db198251d2ed33be2ced8969e6f6f7dfb9a411ce71183278c
-size 387990
+oid sha256:b5ce7b3c6b25e798d2cfa8e192c7fd48d40f2b18556997b0635528190b3c971b
+size 390108