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": 75567.64419142313,
-      "required_pages_per_second_upper_estimate": 0.006550011874069355
+      "required_token_positions_per_second": 75728.04004433253,
+      "required_pages_per_second_upper_estimate": 0.00656391458007997
     },
     {
       "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": 9445.95552392789,
-      "required_pages_per_second_upper_estimate": 0.003768132888115482
+      "required_token_positions_per_second": 9466.005005541567,
+      "required_pages_per_second_upper_estimate": 0.0037761309260976412
     },
     {
       "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": 18891.91104785578,
-      "required_pages_per_second_upper_estimate": 0.004370311134642809
+      "required_token_positions_per_second": 18932.010011083134,
+      "required_pages_per_second_upper_estimate": 0.004379587324067807
     },
     {
       "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": 28337.86657178367,
-      "required_pages_per_second_upper_estimate": 0.0017256273969261102
+      "required_token_positions_per_second": 28398.0150166247,
+      "required_pages_per_second_upper_estimate": 0.0017292901216424278
     },
     {
       "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": 28337.86657178367,
-      "required_pages_per_second_upper_estimate": 0.002181527412931232
+      "required_token_positions_per_second": 28398.0150166247,
+      "required_pages_per_second_upper_estimate": 0.0021861578067166464
     },
     {
       "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": 28337.86657178367,
-      "required_pages_per_second_upper_estimate": 0.3518614620830633
+      "required_token_positions_per_second": 28398.0150166247,
+      "required_pages_per_second_upper_estimate": 0.3526083044640935
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10868897265601163,
-    "producer_p95_seconds": 0.34813446900079725,
-    "gpu_update_seconds": 0.9756556630670786,
-    "producer_batches_per_second": 9.200565389139223,
-    "gpu_batches_per_second": 1.0249517712595366,
-    "producer_headroom": 8.976583725332643,
+    "producer_batch_seconds": 0.13681584352349319,
+    "producer_p95_seconds": 0.3662117389976629,
+    "gpu_update_seconds": 0.9735891745889412,
+    "producer_batches_per_second": 7.309095015945913,
+    "gpu_batches_per_second": 1.0271272792471318,
+    "producer_headroom": 7.116055783566926,
     "sustainable": true,
-    "buffer_seconds": 60.49065111015887,
+    "buffer_seconds": 60.36252882451436,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 341753.4634517087,
-    "upload_bytes_per_second": 6270802.090323461,
+    "generation_bytes_per_second": 343127.70363649057,
+    "upload_bytes_per_second": 6186195.872187354,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.33233764327476,
+    "estimated_full_shards_per_hour": 73.62721759625471,
     "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": 0.900414362000447,
-      "bytes": 7293854,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.0943440850023762,
       "bytes": 1102979,
@@ -562,6 +556,12 @@
       "bytes": 6329429,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.009647611001128,
+      "bytes": 2308548,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -302,3 +302,4 @@
 {"step": 30200, "loss": 0.6555165983736515, "lr": 0.0003, "grad_norm": 0.1274789077023394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4419506132, "tokens_per_second": 150744.9750210262, "data_wait_s": 1.0255997040076181e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 30300, "loss": 0.6104271803051233, "lr": 0.0003, "grad_norm": 0.14621686984475263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4434240188, "tokens_per_second": 149897.92644218702, "data_wait_s": 1.0044997907243669e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 30400, "loss": 0.6053242683410645, "lr": 0.0003, "grad_norm": 0.139099848795369, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4448848223, "tokens_per_second": 149156.09676159924, "data_wait_s": 1.3413999113254249e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 30500, "loss": 0.625363489612937, "lr": 0.0003, "grad_norm": 0.13734782347683128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4463531127, "tokens_per_second": 149835.55554351117, "data_wait_s": 1.020600029733032e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0fe0466c8ee34b2dc3bd995b84cb57da632ad66c2ed2a12b8c953f3a3e2ec3d7
-size 642961
+oid sha256:3abd047739eb217f81b1e4a475dec548e2d10a2bf4f94e9bda731b06d60166d6
+size 645079