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": 76632.04882138042,
-      "required_pages_per_second_upper_estimate": 0.005301582948313918
+      "required_token_positions_per_second": 76787.43464351166,
+      "required_pages_per_second_upper_estimate": 0.005312332900033737
     },
     {
       "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": 9579.006102672553,
-      "required_pages_per_second_upper_estimate": 0.004010529822795132
+      "required_token_positions_per_second": 9598.429330438958,
+      "required_pages_per_second_upper_estimate": 0.0040186619226578076
     },
     {
       "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": 19158.012205345105,
-      "required_pages_per_second_upper_estimate": 0.003990108711616234
+      "required_token_positions_per_second": 19196.858660877915,
+      "required_pages_per_second_upper_estimate": 0.003998199403853794
     },
     {
       "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": 28737.01830801766,
-      "required_pages_per_second_upper_estimate": 0.0017227403686793724
+      "required_token_positions_per_second": 28795.287991316876,
+      "required_pages_per_second_upper_estimate": 0.0017262335472205056
     },
     {
       "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": 28737.01830801766,
-      "required_pages_per_second_upper_estimate": 0.002234176772663283
+      "required_token_positions_per_second": 28795.287991316876,
+      "required_pages_per_second_upper_estimate": 0.002238706984238547
     },
     {
       "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": 28737.01830801766,
-      "required_pages_per_second_upper_estimate": 0.3686076154489766
+      "required_token_positions_per_second": 28795.287991316876,
+      "required_pages_per_second_upper_estimate": 0.36935503638122746
     }
   ],
   "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": 0.2621738818515382,
-    "producer_p95_seconds": 0.3253978350039688,
-    "gpu_update_seconds": 0.9621039908752878,
-    "producer_batches_per_second": 3.814262477016197,
-    "gpu_batches_per_second": 1.039388683015685,
-    "producer_headroom": 3.6697171513831437,
+    "producer_batch_seconds": 0.1051529105860709,
+    "producer_p95_seconds": 0.3165946279987111,
+    "gpu_update_seconds": 0.9601570926582559,
+    "producer_batches_per_second": 9.509960251470824,
+    "gpu_batches_per_second": 1.0414962381118662,
+    "producer_headroom": 9.131055786347803,
     "sustainable": true,
-    "buffer_seconds": 60.61255142514313,
+    "buffer_seconds": 60.48989683747012,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
     "upload_pages_per_commit": 157,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 686526.7018945889,
+    "generation_bytes_per_second": 685479.3678284434,
     "upload_bytes_per_second": 5529363.7270124415,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.65632435144543,
+    "estimated_full_shards_per_hour": 73.54395759649266,
     "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-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -349,3 +349,4 @@
 {"step": 679900, "loss": 0.5150192379951477, "lr": 0.0001, "grad_norm": 0.19555169141593992, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99519707470, "tokens_per_second": 150709.15686407886, "data_wait_s": 9.866002073977143e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 680000, "loss": 0.5438432693481445, "lr": 0.0001, "grad_norm": 0.20167725899899225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99534191298, "tokens_per_second": 150871.13931482463, "data_wait_s": 9.87599923973903e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 680100, "loss": 0.5782261490821838, "lr": 0.0001, "grad_norm": 0.20099426032280923, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99548958680, "tokens_per_second": 141653.32073569723, "data_wait_s": 1.4049997844267637e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 680200, "loss": 0.5280165672302246, "lr": 0.0001, "grad_norm": 0.2234420342974073, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99563650571, "tokens_per_second": 153044.18241708132, "data_wait_s": 1.1339005141053349e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c3b7ecfb4f519d73b1637d90155aeea2cfc82467c193cc8c24b0df6efa10647a
-size 749064
+oid sha256:28cd46daaea66dc8d6777d5a3de9a72144245a0c1cfd1ca586bf0db41001f736
+size 751182