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": 76476.03866312363,
-      "required_pages_per_second_upper_estimate": 0.005290789803572273
+      "required_token_positions_per_second": 77010.54843092566,
+      "required_pages_per_second_upper_estimate": 0.005327768429542349
     },
     {
       "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": 9559.504832890454,
-      "required_pages_per_second_upper_estimate": 0.004002365048370189
+      "required_token_positions_per_second": 9626.318553865707,
+      "required_pages_per_second_upper_estimate": 0.004030338558113376
     },
     {
       "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": 19119.009665780908,
-      "required_pages_per_second_upper_estimate": 0.003981985511191148
+      "required_token_positions_per_second": 19252.637107731414,
+      "required_pages_per_second_upper_estimate": 0.004009816583356815
     },
     {
       "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": 28678.514498671364,
-      "required_pages_per_second_upper_estimate": 0.0017192331546392063
+      "required_token_positions_per_second": 28878.95566159712,
+      "required_pages_per_second_upper_estimate": 0.0017312492963006709
     },
     {
       "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": 28678.514498671364,
-      "required_pages_per_second_upper_estimate": 0.00222962835881767
+      "required_token_positions_per_second": 28878.95566159712,
+      "required_pages_per_second_upper_estimate": 0.002245211777587647
     },
     {
       "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": 28678.514498671364,
-      "required_pages_per_second_upper_estimate": 0.3678571913991786
+      "required_token_positions_per_second": 28878.95566159712,
+      "required_pages_per_second_upper_estimate": 0.3704282354202373
     }
   ],
   "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": 0.8526361567656409,
-    "producer_p95_seconds": 0.3472557479981333,
-    "gpu_update_seconds": 0.9640666709316793,
-    "producer_batches_per_second": 1.172833209177246,
-    "gpu_batches_per_second": 1.0372726598188426,
-    "producer_headroom": 1.1306894075296254,
+    "producer_batch_seconds": 0.11551204424984007,
+    "producer_p95_seconds": 0.32795128100406146,
+    "gpu_update_seconds": 0.9573753401604467,
+    "producer_batches_per_second": 8.657105901763005,
+    "gpu_batches_per_second": 1.044522412528831,
+    "producer_headroom": 8.288099707505369,
     "sustainable": true,
-    "buffer_seconds": 60.7362002686958,
+    "buffer_seconds": 60.31464643010814,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 265600.2828675304,
+    "generation_bytes_per_second": 265978.39310200425,
     "upload_bytes_per_second": 4087731.5521492357,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.99163784522471,
+    "estimated_full_shards_per_hour": 57.07277149958702,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -349,3 +349,4 @@
 {"step": 877400, "loss": 0.5608643293380737, "lr": 3e-05, "grad_norm": 0.32331489554311, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128427011320, "tokens_per_second": 150429.5205340763, "data_wait_s": 8.13999940874055e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 877500, "loss": 0.49823275208473206, "lr": 3e-05, "grad_norm": 0.3163390680099127, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128441726918, "tokens_per_second": 119242.43539023101, "data_wait_s": 1.7575999663677067e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 877600, "loss": 0.5132738947868347, "lr": 3e-05, "grad_norm": 0.34010128533589473, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128456356294, "tokens_per_second": 151670.18337014448, "data_wait_s": 1.408400567015633e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 877700, "loss": 0.5507510900497437, "lr": 3e-05, "grad_norm": 0.3248438304098416, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128471066757, "tokens_per_second": 153429.1232770866, "data_wait_s": 9.777002560440451e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c0e58ee595d3b18defa0a74aaba96e80a63e592973bf759e684fa795842efd29
-size 749064
+oid sha256:8c8b808faf402d3124a0a49a82550ec09e998cb3292af39361020e8e61d1cc26
+size 751182