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": 75370.39799946047,
-      "required_pages_per_second_upper_estimate": 0.0052142990170202675
+      "required_token_positions_per_second": 75783.76696073382,
+      "required_pages_per_second_upper_estimate": 0.005242896840909292
     },
     {
       "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": 9421.29974993256,
-      "required_pages_per_second_upper_estimate": 0.003944501466186034
+      "required_token_positions_per_second": 9472.970870091727,
+      "required_pages_per_second_upper_estimate": 0.00396613508518099
     },
     {
       "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": 18842.59949986512,
-      "required_pages_per_second_upper_estimate": 0.0039244165630571565
+      "required_token_positions_per_second": 18945.941740183454,
+      "required_pages_per_second_upper_estimate": 0.0039459400263973194
     },
     {
       "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": 28263.89924979768,
-      "required_pages_per_second_upper_estimate": 0.0016943776035500575
+      "required_token_positions_per_second": 28418.91261027518,
+      "required_pages_per_second_upper_estimate": 0.001703670417818987
     },
     {
       "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": 28263.89924979768,
-      "required_pages_per_second_upper_estimate": 0.0021973938469174864
+      "required_token_positions_per_second": 28418.91261027518,
+      "required_pages_per_second_upper_estimate": 0.0022094454538628973
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28263.89924979768,
-      "required_pages_per_second_upper_estimate": 0.36253895216579674
+      "required_token_positions_per_second": 28418.91261027518,
+      "required_pages_per_second_upper_estimate": 0.3645272971136232
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.09402027436718186,
-    "producer_p95_seconds": 0.26729268999770284,
-    "gpu_update_seconds": 0.978208978019829,
-    "producer_batches_per_second": 10.636003848432226,
-    "gpu_batches_per_second": 1.0222764485603906,
-    "producer_headroom": 10.404234454789856,
+    "producer_batch_seconds": 0.09700551107007982,
+    "producer_p95_seconds": 0.26210722399991937,
+    "gpu_update_seconds": 0.9728732544820716,
+    "producer_batches_per_second": 10.308692660539345,
+    "gpu_batches_per_second": 1.0278831239248836,
+    "producer_headroom": 10.029051378114358,
     "sustainable": true,
-    "buffer_seconds": 60.6489566372294,
+    "buffer_seconds": 60.31814177788844,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 378558.35604428645,
+    "generation_bytes_per_second": 380025.4138056223,
     "upload_bytes_per_second": 3865438.6318284855,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.22981081959195,
+    "estimated_full_shards_per_hour": 81.54460726381781,
     "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

@@ -263,3 +263,4 @@
 {"step": 551300, "loss": 0.588507890701294, "lr": 0.0003, "grad_norm": 0.1714966714367206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80692935498, "tokens_per_second": 153859.97673126977, "data_wait_s": 1.3256001693662256e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 551400, "loss": 0.6358245015144348, "lr": 0.0003, "grad_norm": 0.17163011160037733, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80707636723, "tokens_per_second": 152768.05945944792, "data_wait_s": 2.1840998670086265e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 551500, "loss": 0.5715614557266235, "lr": 0.0003, "grad_norm": 0.1479404510899116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80722270761, "tokens_per_second": 152463.46489181934, "data_wait_s": 8.75200203154236e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 551600, "loss": 0.5878363847732544, "lr": 0.0003, "grad_norm": 0.17073314825640917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80736882669, "tokens_per_second": 152311.14371276874, "data_wait_s": 1.499599966336973e-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 sha256:5e08b189f7c06b7f2d3ac7804efa7b74a0f70f43b9ab1c42c5e7b5a2e14278bb
-size 565350
+oid sha256:c71e381f845613084679249addf7a40db89449a012c62489d497d33c231831c8
+size 567468