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": 75728.04004433253,
-      "required_pages_per_second_upper_estimate": 0.00656391458007997
+      "required_token_positions_per_second": 75670.76386564385,
+      "required_pages_per_second_upper_estimate": 0.006558950026076382
     },
     {
       "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": 9466.005005541567,
-      "required_pages_per_second_upper_estimate": 0.0037761309260976412
+      "required_token_positions_per_second": 9458.845483205481,
+      "required_pages_per_second_upper_estimate": 0.0037732748855933787
     },
     {
       "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": 18932.010011083134,
-      "required_pages_per_second_upper_estimate": 0.004379587324067807
+      "required_token_positions_per_second": 18917.690966410963,
+      "required_pages_per_second_upper_estimate": 0.00437627486508948
     },
     {
       "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": 28398.0150166247,
-      "required_pages_per_second_upper_estimate": 0.0017292901216424278
+      "required_token_positions_per_second": 28376.53644961644,
+      "required_pages_per_second_upper_estimate": 0.0017279821896009568
     },
     {
       "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": 28398.0150166247,
-      "required_pages_per_second_upper_estimate": 0.0021861578067166464
+      "required_token_positions_per_second": 28376.53644961644,
+      "required_pages_per_second_upper_estimate": 0.002184504327171872
     },
     {
       "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": 28398.0150166247,
-      "required_pages_per_second_upper_estimate": 0.3526083044640935
+      "required_token_positions_per_second": 28376.53644961644,
+      "required_pages_per_second_upper_estimate": 0.35234161254599056
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "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,
+    "producer_batch_seconds": 0.10633590773412038,
+    "producer_p95_seconds": 0.37021873999765376,
+    "gpu_update_seconds": 0.9743260968120622,
+    "producer_batches_per_second": 9.404161033734482,
+    "gpu_batches_per_second": 1.0263504213547614,
+    "producer_headroom": 9.162719513790606,
     "sustainable": true,
-    "buffer_seconds": 60.36252882451436,
+    "buffer_seconds": 60.40821800234785,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 343127.70363649057,
+    "generation_bytes_per_second": 302016.2642299256,
     "upload_bytes_per_second": 6186195.872187354,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.62721759625471,
+    "estimated_full_shards_per_hour": 64.80565972493483,
     "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": [

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

@@ -303,3 +303,4 @@
 {"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}
+{"step": 30600, "loss": 0.6761391311883926, "lr": 0.0003, "grad_norm": 0.12689351908525573, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4477997788, "tokens_per_second": 147090.327656845, "data_wait_s": 1.707800038275309e-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 sha256:3abd047739eb217f81b1e4a475dec548e2d10a2bf4f94e9bda731b06d60166d6
-size 645079
+oid sha256:7b0ffa5086f9ed33624eee47431203092b03f493a89aac801ee6f2859da648de
+size 647197