Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75853.18986298314,
-      "required_pages_per_second_upper_estimate": 0.006574762249171289
+      "required_token_positions_per_second": 76019.88883724359,
+      "required_pages_per_second_upper_estimate": 0.006589211293765497
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9481.648732872893,
-      "required_pages_per_second_upper_estimate": 0.003782371442824674
+      "required_token_positions_per_second": 9502.486104655449,
+      "required_pages_per_second_upper_estimate": 0.003790683781975207
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18963.297465745785,
-      "required_pages_per_second_upper_estimate": 0.004386825126063617
+      "required_token_positions_per_second": 19004.972209310898,
+      "required_pages_per_second_upper_estimate": 0.0043964658445369694
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28444.94619861868,
-      "required_pages_per_second_upper_estimate": 0.0017321479843969787
+      "required_token_positions_per_second": 28507.45831396635,
+      "required_pages_per_second_upper_estimate": 0.0017359546442459299
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28444.94619861868,
-      "required_pages_per_second_upper_estimate": 0.0021897706990203695
+      "required_token_positions_per_second": 28507.45831396635,
+      "required_pages_per_second_upper_estimate": 0.0021945830547044436
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28444.94619861868,
-      "required_pages_per_second_upper_estimate": 0.3531910326758965
+      "required_token_positions_per_second": 28507.45831396635,
+      "required_pages_per_second_upper_estimate": 0.3539672239339229
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.13037884117952103,
-    "producer_p95_seconds": 0.3595136559997627,
-    "gpu_update_seconds": 0.9719828544215218,
-    "producer_batches_per_second": 7.669956190384309,
-    "gpu_batches_per_second": 1.028824732299576,
-    "producer_headroom": 7.455065911217761,
+    "producer_batch_seconds": 0.11164077478872514,
+    "producer_p95_seconds": 0.35594936199777294,
+    "gpu_update_seconds": 0.9698514576606332,
+    "producer_batches_per_second": 8.957300788107682,
+    "gpu_batches_per_second": 1.0310857318419542,
+    "producer_headroom": 8.687251226050975,
     "sustainable": true,
-    "buffer_seconds": 60.26293697413435,
+    "buffer_seconds": 60.13079037495926,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360852.849047202,
+    "generation_bytes_per_second": 360967.69278708415,
     "upload_bytes_per_second": 6226915.286975836,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.43062118112607,
+    "estimated_full_shards_per_hour": 77.45526397427933,
     "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-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -280,3 +280,4 @@
 {"step": 28000, "loss": 0.7334572561085224, "lr": 0.0003, "grad_norm": 0.16861153780658747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4097369174, "tokens_per_second": 139404.03716623955, "data_wait_s": 1.0496001777937636e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 28100, "loss": 0.6417201161384583, "lr": 0.0003, "grad_norm": 0.13232709154429506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4112073984, "tokens_per_second": 150423.2972179369, "data_wait_s": 8.861999958753586e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 28200, "loss": 0.6506574675440788, "lr": 0.0003, "grad_norm": 0.1307841212003055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4126800639, "tokens_per_second": 149813.64561755268, "data_wait_s": 1.1559000995475799e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 28300, "loss": 0.680478785187006, "lr": 0.0003, "grad_norm": 0.1456216874475559, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4141315645, "tokens_per_second": 148029.4598356252, "data_wait_s": 1.1807002010755241e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:947f2ed0b73f8d5a8110584721489229987babaf7875eb507bbc437632766cbc
-size 595582
+oid sha256:71469517559c8b695361018b7ac4d18968f92f26816b664bf9320f95026ee898
+size 597700