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": 77012.862282268,
-      "required_pages_per_second_upper_estimate": 0.005327928507147087
+      "required_token_positions_per_second": 77201.67287745554,
+      "required_pages_per_second_upper_estimate": 0.0053409908622231005
     },
     {
       "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": 9626.6077852835,
-      "required_pages_per_second_upper_estimate": 0.004030459653268168
+      "required_token_positions_per_second": 9650.209109681942,
+      "required_pages_per_second_upper_estimate": 0.004040341034942097
     },
     {
       "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": 19253.215570567,
-      "required_pages_per_second_upper_estimate": 0.004009937061910378
+      "required_token_positions_per_second": 19300.418219363884,
+      "required_pages_per_second_upper_estimate": 0.004019768128837209
     },
     {
       "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": 28879.823355850505,
-      "required_pages_per_second_upper_estimate": 0.001731301313246269
+      "required_token_positions_per_second": 28950.627329045827,
+      "required_pages_per_second_upper_estimate": 0.0017355459033279212
     },
     {
       "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": 28879.823355850505,
-      "required_pages_per_second_upper_estimate": 0.002245279236999264
+      "required_token_positions_per_second": 28950.627329045827,
+      "required_pages_per_second_upper_estimate": 0.002250783934481436
     },
     {
       "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": 28879.823355850505,
-      "required_pages_per_second_upper_estimate": 0.37043936527046223
+      "required_token_positions_per_second": 28950.627329045827,
+      "required_pages_per_second_upper_estimate": 0.37134756261522844
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12190655599977163,
-    "producer_p95_seconds": 0.325343017000705,
-    "gpu_update_seconds": 0.9573465758196558,
-    "producer_batches_per_second": 8.20300427486339,
-    "gpu_batches_per_second": 1.0445537961462132,
-    "producer_headroom": 7.853118053974465,
+    "producer_batch_seconds": 0.4964218595780494,
+    "producer_p95_seconds": 0.34940056299819844,
+    "gpu_update_seconds": 0.9550052123485796,
+    "producer_batches_per_second": 2.014415724662052,
+    "gpu_batches_per_second": 1.04711470374153,
+    "producer_headroom": 1.9237775168892013,
     "sustainable": true,
-    "buffer_seconds": 60.31283427663831,
+    "buffer_seconds": 60.16532837796051,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 304681.1501766358,
-    "upload_bytes_per_second": 4326037.661222872,
+    "generation_bytes_per_second": 292058.9534759953,
+    "upload_bytes_per_second": 4236218.212568838,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.37748221372897,
+    "estimated_full_shards_per_hour": 62.66905263147253,
     "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": 1.546362834997126,
-      "bytes": 11822887,
-      "pages": 33,
-      "shards": 4
-    },
     {
       "seconds": 1.5667999549987144,
       "bytes": 6923893,
@@ -562,6 +556,12 @@
       "bytes": 11835967,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.3631727270039846,
+      "bytes": 2146631,
+      "pages": 21,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -407,3 +407,4 @@
 {"step": 840700, "loss": 0.5038002729415894, "lr": 3e-05, "grad_norm": 0.2786213606761156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123054808705, "tokens_per_second": 151234.74413240745, "data_wait_s": 1.2336000509094447e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 840800, "loss": 0.5980193018913269, "lr": 3e-05, "grad_norm": 0.32692752093060673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123069241731, "tokens_per_second": 150399.62103053424, "data_wait_s": 1.2342003174126148e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 840900, "loss": 0.5097891688346863, "lr": 3e-05, "grad_norm": 0.28425541977736746, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123083732760, "tokens_per_second": 133410.07703303223, "data_wait_s": 9.864997991826385e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 841000, "loss": 0.5003107190132141, "lr": 3e-05, "grad_norm": 0.3166021561959973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123098235734, "tokens_per_second": 150518.3236412449, "data_wait_s": 9.83799691312015e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2d85da3ad265a620b33a1d6d33a9159c8834b432f85e58a11cd772538277761c
-size 872691
+oid sha256:84b36f598b4a1024cca1eac2a036e7a441b73e92f4ca8807c6e9bae01bf48bbb
+size 874809