Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76491.62459411804,
-      "required_pages_per_second_upper_estimate": 0.005291868074442809
+      "required_token_positions_per_second": 76868.88460164159,
+      "required_pages_per_second_upper_estimate": 0.005317967797127106
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9561.453074264755,
-      "required_pages_per_second_upper_estimate": 0.004003180736349702
+      "required_token_positions_per_second": 9608.610575205199,
+      "required_pages_per_second_upper_estimate": 0.004022924597232865
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19122.90614852951,
-      "required_pages_per_second_upper_estimate": 0.003982797045790521
+      "required_token_positions_per_second": 19217.221150410398,
+      "required_pages_per_second_upper_estimate": 0.00400244037342845
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28684.359222794264,
-      "required_pages_per_second_upper_estimate": 0.0017195835369259972
+      "required_token_positions_per_second": 28825.831725615597,
+      "required_pages_per_second_upper_estimate": 0.001728064597454132
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28684.359222794264,
-      "required_pages_per_second_upper_estimate": 0.002230082760410001
+      "required_token_positions_per_second": 28825.831725615597,
+      "required_pages_per_second_upper_estimate": 0.0022410816252325834
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28684.359222794264,
-      "required_pages_per_second_upper_estimate": 0.36793216124465133
+      "required_token_positions_per_second": 28825.831725615597,
+      "required_pages_per_second_upper_estimate": 0.3697468186095047
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11137248615622752,
-    "producer_p95_seconds": 0.17949492200023087,
-    "gpu_update_seconds": 0.9638702327374734,
-    "producer_batches_per_second": 8.978878307495552,
-    "gpu_batches_per_second": 1.0374840575374082,
-    "producer_headroom": 8.654473523967189,
+    "producer_batch_seconds": 0.40052486080468697,
+    "producer_p95_seconds": 0.1698077959999864,
+    "gpu_update_seconds": 0.9591397140999427,
+    "producer_batches_per_second": 2.4967239186873913,
+    "gpu_batches_per_second": 1.0426009738720918,
+    "producer_headroom": 2.394707065556313,
     "sustainable": true,
-    "buffer_seconds": 60.72382466246083,
+    "buffer_seconds": 60.42580198829639,
     "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": 337574.35614757927,
-    "upload_bytes_per_second": 4781195.518718916,
+    "generation_bytes_per_second": 307592.2273997774,
+    "upload_bytes_per_second": 4833616.798351569,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.4355985004476,
+    "estimated_full_shards_per_hour": 66.00213161940566,
     "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": [
@@ -310,6 +310,12 @@
       "bytes": 6761235,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.2538779149999755,
+      "bytes": 7546399,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -66,3 +66,4 @@
 {"step": 726600, "loss": 0.5636150240898132, "lr": 5e-05, "grad_norm": 0.2252468873031572, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106353808082, "tokens_per_second": 150484.9628767221, "data_wait_s": 1.3204000424593687e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 726700, "loss": 0.5367215275764465, "lr": 5e-05, "grad_norm": 0.2096011994271188, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106368390154, "tokens_per_second": 150624.68540030325, "data_wait_s": 1.270099983230466e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 726800, "loss": 0.5006130337715149, "lr": 5e-05, "grad_norm": 0.22042341033664942, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106383099106, "tokens_per_second": 134237.54735743703, "data_wait_s": 1.0092000593431294e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 726900, "loss": 0.5659409165382385, "lr": 5e-05, "grad_norm": 0.23362119149740934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106397726764, "tokens_per_second": 151721.40579858958, "data_wait_s": 1.1424999684095383e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e3dd494b4f59a52f8cedd16c0d1843c3f454f7087db9669959300a35d6b0926b
-size 144972
+oid sha256:ff29bb2eb560ea5b32a5fa74637b5da302cb8e0acdc693a610822171a73877c8
+size 147090