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": 77042.14783194858,
-      "required_pages_per_second_upper_estimate": 0.0053299545494258265
+      "required_token_positions_per_second": 76926.41189750822,
+      "required_pages_per_second_upper_estimate": 0.005321947668936877
     },
     {
       "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": 9630.268478993572,
-      "required_pages_per_second_upper_estimate": 0.004031992309280597
+      "required_token_positions_per_second": 9615.801487188528,
+      "required_pages_per_second_upper_estimate": 0.004025935281916968
     },
     {
       "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": 19260.536957987144,
-      "required_pages_per_second_upper_estimate": 0.004011461913832023
+      "required_token_positions_per_second": 19231.602974377056,
+      "required_pages_per_second_upper_estimate": 0.004005435728086502
     },
     {
       "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": 28890.805436980718,
-      "required_pages_per_second_upper_estimate": 0.0017319596722413582
+      "required_token_positions_per_second": 28847.404461565588,
+      "required_pages_per_second_upper_estimate": 0.0017293578500346735
     },
     {
       "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": 28890.805436980718,
-      "required_pages_per_second_upper_estimate": 0.0022461330455020687
+      "required_token_positions_per_second": 28847.404461565588,
+      "required_pages_per_second_upper_estimate": 0.002242758810564258
     },
     {
       "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": 28890.805436980718,
-      "required_pages_per_second_upper_estimate": 0.3705802316155606
+      "required_token_positions_per_second": 28847.404461565588,
+      "required_pages_per_second_upper_estimate": 0.37002353050327197
     }
   ],
   "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": 0.1497711076405892,
-    "producer_p95_seconds": 0.19726968699978897,
-    "gpu_update_seconds": 0.9569826656549386,
-    "producer_batches_per_second": 6.676855207612765,
-    "gpu_batches_per_second": 1.0449510068352401,
-    "producer_headroom": 6.3896346947733225,
+    "producer_batch_seconds": 0.07830520684377973,
+    "producer_p95_seconds": 0.2024847860011505,
+    "gpu_update_seconds": 0.9584224479133436,
+    "producer_batches_per_second": 12.770542857961127,
+    "gpu_batches_per_second": 1.043381237759172,
+    "producer_headroom": 12.239574947109372,
     "sustainable": true,
-    "buffer_seconds": 60.28990793626113,
+    "buffer_seconds": 60.38061421854065,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 401552.4931590582,
+    "generation_bytes_per_second": 403373.9589439112,
     "upload_bytes_per_second": 5095392.955881391,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.1638173683053,
+    "estimated_full_shards_per_hour": 86.55466152418138,
     "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-v1/1790868774355361490-de221e1d/train.jsonl

@@ -131,3 +131,4 @@
 {"step": 453300, "loss": 0.593574583530426, "lr": 0.0003, "grad_norm": 0.14305193484302964, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66349042421, "tokens_per_second": 153855.3941027908, "data_wait_s": 1.2913998943986371e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 453400, "loss": 0.6120077967643738, "lr": 0.0003, "grad_norm": 0.16437432505629354, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66363950376, "tokens_per_second": 155389.68197901294, "data_wait_s": 1.4379000276676379e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 453500, "loss": 0.5911997556686401, "lr": 0.0003, "grad_norm": 0.19097494869260073, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66378659727, "tokens_per_second": 153963.6426333046, "data_wait_s": 1.2004999007331207e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 453600, "loss": 0.5883230566978455, "lr": 0.0003, "grad_norm": 0.15073408393150328, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66393277437, "tokens_per_second": 152126.32827518907, "data_wait_s": 1.2428999980329536e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:058fbefac635797a09f71e442037cad4f55edf36d00894932d43423503998fc8
-size 283425
+oid sha256:b0288b69b2f1f13110b781466580c616e6a01ac959f4a85d9021df8edae56771
+size 285543