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": 76773.42676630623,
-      "required_pages_per_second_upper_estimate": 0.005311363802585913
+      "required_token_positions_per_second": 76882.74181453539,
+      "required_pages_per_second_upper_estimate": 0.005318926471268263
     },
     {
       "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": 9596.67834578828,
-      "required_pages_per_second_upper_estimate": 0.004017928821949286
+      "required_token_positions_per_second": 9610.342726816923,
+      "required_pages_per_second_upper_estimate": 0.004023649812941256
     },
     {
       "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": 19193.35669157656,
-      "required_pages_per_second_upper_estimate": 0.003997470036001463
+      "required_token_positions_per_second": 19220.685453633847,
+      "required_pages_per_second_upper_estimate": 0.004003161896430075
     },
     {
       "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": 28790.035037364836,
-      "required_pages_per_second_upper_estimate": 0.0017259186406518771
+      "required_token_positions_per_second": 28831.028180450772,
+      "required_pages_per_second_upper_estimate": 0.0017283761169869737
     },
     {
       "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": 28790.035037364836,
-      "required_pages_per_second_upper_estimate": 0.002238298590174078
+      "required_token_positions_per_second": 28831.028180450772,
+      "required_pages_per_second_upper_estimate": 0.002241485626739229
     },
     {
       "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": 28790.035037364836,
-      "required_pages_per_second_upper_estimate": 0.36928765712811323
+      "required_token_positions_per_second": 28831.028180450772,
+      "required_pages_per_second_upper_estimate": 0.36981347315261187
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.07441354914843146,
-    "producer_p95_seconds": 0.17217312999946444,
-    "gpu_update_seconds": 0.9603322803920642,
-    "producer_batches_per_second": 13.438412915977395,
-    "gpu_batches_per_second": 1.0413062441176517,
-    "producer_headroom": 12.90534172045074,
+    "producer_batch_seconds": 0.06674609317187219,
+    "producer_p95_seconds": 0.18033385900071153,
+    "gpu_update_seconds": 0.9589668404107441,
+    "producer_batches_per_second": 14.98215030241523,
+    "gpu_batches_per_second": 1.0427889243507946,
+    "producer_headroom": 14.367385338066006,
     "sustainable": true,
-    "buffer_seconds": 60.50093366470004,
+    "buffer_seconds": 60.414910945876876,
     "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": 374457.928429165,
+    "generation_bytes_per_second": 391406.95234401204,
     "upload_bytes_per_second": 3937205.9001927166,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.34995450645648,
+    "estimated_full_shards_per_hour": 83.98682048549911,
     "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-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -59,3 +59,4 @@
 {"step": 530900, "loss": 0.6229141354560852, "lr": 0.0003, "grad_norm": 0.15822727428106462, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77706350894, "tokens_per_second": 154314.1423603472, "data_wait_s": 1.3328000022738706e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 531000, "loss": 0.5705332159996033, "lr": 0.0003, "grad_norm": 0.1550770650128346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77721076278, "tokens_per_second": 154057.69925950665, "data_wait_s": 1.1037999684049282e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 531100, "loss": 0.555992066860199, "lr": 0.0003, "grad_norm": 0.1407968699878935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77735763717, "tokens_per_second": 153614.5171548599, "data_wait_s": 8.940000043367036e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 531200, "loss": 0.5846951603889465, "lr": 0.0003, "grad_norm": 0.16805864125753955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77750483592, "tokens_per_second": 153595.62734738825, "data_wait_s": 1.3156999557395466e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:16e1402097122ce73bd0b4af1a732202fbdc850d20ddbbda7d6e0e8371b56da4
-size 130146
+oid sha256:b6ed8644bcd052b6bedd6451ad7768710cac5ca48e0bcd373bf258fea11d733a
+size 132264