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": 74961.56090393316,
-      "required_pages_per_second_upper_estimate": 0.006497478110813773
+      "required_token_positions_per_second": 75275.64651541911,
+      "required_pages_per_second_upper_estimate": 0.006524702255574671
     },
     {
       "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": 9370.195112991645,
-      "required_pages_per_second_upper_estimate": 0.003737910927473929
+      "required_token_positions_per_second": 9409.455814427389,
+      "required_pages_per_second_upper_estimate": 0.0037535726082764437
     },
     {
       "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": 18740.39022598329,
-      "required_pages_per_second_upper_estimate": 0.004335259459177995
+      "required_token_positions_per_second": 18818.911628854778,
+      "required_pages_per_second_upper_estimate": 0.004353424003802822
     },
     {
       "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": 28110.585338974935,
-      "required_pages_per_second_upper_estimate": 0.0017117871623005291
+      "required_token_positions_per_second": 28228.367443282168,
+      "required_pages_per_second_upper_estimate": 0.0017189594744978964
     },
     {
       "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": 28110.585338974935,
-      "required_pages_per_second_upper_estimate": 0.0021640306744748956
+      "required_token_positions_per_second": 28228.367443282168,
+      "required_pages_per_second_upper_estimate": 0.002173097866906199
     },
     {
       "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": 28110.585338974935,
-      "required_pages_per_second_upper_estimate": 0.34903938983293314
+      "required_token_positions_per_second": 28228.367443282168,
+      "required_pages_per_second_upper_estimate": 0.35050184937708345
     }
   ],
   "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": 0.46963190768764207,
-    "producer_p95_seconds": 0.37708686599944485,
-    "gpu_update_seconds": 0.9835440872754233,
-    "producer_batches_per_second": 2.1293272105887495,
-    "gpu_batches_per_second": 1.0167312405589892,
-    "producer_headroom": 2.094287187849235,
+    "producer_batch_seconds": 0.12524553974247965,
+    "producer_p95_seconds": 0.3457775569986552,
+    "gpu_update_seconds": 0.9794402760114176,
+    "producer_batches_per_second": 7.9843162643246535,
+    "gpu_batches_per_second": 1.0209912993085275,
+    "producer_headroom": 7.820160925692589,
     "sustainable": true,
-    "buffer_seconds": 60.979733411076246,
+    "buffer_seconds": 60.72529711270789,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337363.02287753817,
-    "upload_bytes_per_second": 4187565.041897179,
+    "generation_bytes_per_second": 337527.1920180909,
+    "upload_bytes_per_second": 3222427.5878580376,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.39025130028351,
+    "estimated_full_shards_per_hour": 72.42547817618414,
     "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": 1.1360236319997057,
-      "bytes": 2731200,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.413960779998888,
       "bytes": 11724293,
@@ -562,6 +556,12 @@
       "bytes": 7005197,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 34.0496005649984,
+      "bytes": 11526687,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -311,3 +311,4 @@
 {"step": 346100, "loss": 0.607675364241004, "lr": 0.0003, "grad_norm": 0.16171144785741007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50660075345, "tokens_per_second": 149965.76371127076, "data_wait_s": 1.42700009746477e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 346200, "loss": 0.5823972150683403, "lr": 0.0003, "grad_norm": 0.14689265943939842, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50674792131, "tokens_per_second": 149689.65688336254, "data_wait_s": 1.903999873320572e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 346300, "loss": 0.5719314459711313, "lr": 0.0003, "grad_norm": 0.15280105862423843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50689435994, "tokens_per_second": 149228.67381966935, "data_wait_s": 1.6827998479129747e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 346400, "loss": 0.5925719030201435, "lr": 0.0003, "grad_norm": 0.1437341434483271, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50703960052, "tokens_per_second": 147306.37665867622, "data_wait_s": 1.664400042500347e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dc87495a7943c5eb43831502df04a8b5adda2f1aab507e0fb9421aba5b3b4198
-size 667797
+oid sha256:a95fc3d00e1b270c22eee4985279d616c852f04d8a506a072d04829adfec791e
+size 669915