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": 77211.8910411284,
-      "required_pages_per_second_upper_estimate": 0.005341697778495395
+      "required_token_positions_per_second": 77309.42002761336,
+      "required_pages_per_second_upper_estimate": 0.005348445060078852
     },
     {
       "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": 9651.48638014105,
-      "required_pages_per_second_upper_estimate": 0.004040875801410886
+      "required_token_positions_per_second": 9663.67750345167,
+      "required_pages_per_second_upper_estimate": 0.004045979970161439
     },
     {
       "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": 19302.9727602821,
-      "required_pages_per_second_upper_estimate": 0.004020300172342699
+      "required_token_positions_per_second": 19327.35500690334,
+      "required_pages_per_second_upper_estimate": 0.00402537835131082
     },
     {
       "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": 28954.459140423147,
-      "required_pages_per_second_upper_estimate": 0.0017357756145691544
+      "required_token_positions_per_second": 28991.032510355013,
+      "required_pages_per_second_upper_estimate": 0.0017379681322522674
     },
     {
       "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": 28954.459140423147,
-      "required_pages_per_second_upper_estimate": 0.0022510818409616656
+      "required_token_positions_per_second": 28991.032510355013,
+      "required_pages_per_second_upper_estimate": 0.002253925259604615
     },
     {
       "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": 28954.459140423147,
-      "required_pages_per_second_upper_estimate": 0.3713967129772982
+      "required_token_positions_per_second": 28991.032510355013,
+      "required_pages_per_second_upper_estimate": 0.3718658368973591
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.07043672639055387,
-    "producer_p95_seconds": 0.18387975099903997,
-    "gpu_update_seconds": 0.9548788276760035,
-    "producer_batches_per_second": 14.197139067128878,
-    "gpu_batches_per_second": 1.047253296456277,
-    "producer_headroom": 13.556547508773212,
+    "producer_batch_seconds": 0.09876783672650902,
+    "producer_p95_seconds": 0.19724191800014523,
+    "gpu_update_seconds": 0.9536742090894724,
+    "producer_batches_per_second": 10.124753494085619,
+    "gpu_batches_per_second": 1.048576117996058,
+    "producer_headroom": 9.655716280697973,
     "sustainable": true,
-    "buffer_seconds": 60.15736614358822,
+    "buffer_seconds": 60.081475172636765,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 278536.40938426624,
-    "upload_bytes_per_second": 5278741.68078892,
+    "generation_bytes_per_second": 301876.5270565495,
+    "upload_bytes_per_second": 5222573.164601149,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.76742945810309,
+    "estimated_full_shards_per_hour": 64.77567538044323,
     "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": [
@@ -520,6 +520,12 @@
       "bytes": 11967346,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 0.9512228559997311,
+      "bytes": 901267,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -82,3 +82,4 @@
 {"step": 893200, "loss": 0.5528310537338257, "lr": 3e-05, "grad_norm": 0.3245383546079211, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130739254414, "tokens_per_second": 151356.05284035948, "data_wait_s": 1.2751999747706577e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 893300, "loss": 0.5231316089630127, "lr": 3e-05, "grad_norm": 0.31683000040934844, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130753693130, "tokens_per_second": 150829.23747425992, "data_wait_s": 1.5031999282655306e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 893400, "loss": 0.5377864241600037, "lr": 3e-05, "grad_norm": 0.3333041613099153, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130768306219, "tokens_per_second": 152808.71813224693, "data_wait_s": 1.2407001122483052e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 893500, "loss": 0.5325813889503479, "lr": 3e-05, "grad_norm": 0.30968578843609335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130782901508, "tokens_per_second": 135312.5876459826, "data_wait_s": 8.703998901182786e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dbe7fbb8bc38396e5143f154927450104d2e0d910748f35f9d9f0164d893d167
-size 178860
+oid sha256:c4e341e389ce518ee1f1e086495030ff639d55601bdc5691f33aaf34840a1dbc
+size 180978