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": 77631.39846861135,
-      "required_pages_per_second_upper_estimate": 0.005370720275720005
+      "required_token_positions_per_second": 77276.63016099708,
+      "required_pages_per_second_upper_estimate": 0.005346176581023364
     },
     {
       "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": 9703.92480857642,
-      "required_pages_per_second_upper_estimate": 0.004062830676357868
+      "required_token_positions_per_second": 9659.578770124635,
+      "required_pages_per_second_upper_estimate": 0.0040442639161086934
     },
     {
       "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": 19407.84961715284,
-      "required_pages_per_second_upper_estimate": 0.004042143255840167
+      "required_token_positions_per_second": 19319.15754024927,
+      "required_pages_per_second_upper_estimate": 0.0040236710351885105
     },
     {
       "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": 29111.774425729258,
-      "required_pages_per_second_upper_estimate": 0.0017452064257167234
+      "required_token_positions_per_second": 28978.736310373904,
+      "required_pages_per_second_upper_estimate": 0.0017372309938386101
     },
     {
       "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": 29111.774425729258,
-      "required_pages_per_second_upper_estimate": 0.0022633124124374036
+      "required_token_positions_per_second": 28978.736310373904,
+      "required_pages_per_second_upper_estimate": 0.0022529692841413516
     },
     {
       "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": 29111.774425729258,
-      "required_pages_per_second_upper_estimate": 0.3734145845452118
+      "required_token_positions_per_second": 28978.736310373904,
+      "required_pages_per_second_upper_estimate": 0.3717081144466323
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,20 +155,20 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.049732651421866336,
-    "producer_p95_seconds": 0.13528153200013548,
-    "gpu_update_seconds": 0.9497188180863493,
-    "producer_batches_per_second": 20.107514307196627,
-    "gpu_batches_per_second": 1.0529432300972676,
-    "producer_headroom": 19.096484722485137,
+    "producer_batch_seconds": 0.05209850517968917,
+    "producer_p95_seconds": 0.13456797300023027,
+    "gpu_update_seconds": 0.9540788702405383,
+    "producer_batches_per_second": 19.194408679308026,
+    "gpu_batches_per_second": 1.0481313769666487,
+    "producer_headroom": 18.312979747689386,
     "sustainable": true,
-    "buffer_seconds": 60.782004357526354,
+    "buffer_seconds": 60.10696882515391,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -35,3 +35,4 @@
 {"step": 928500, "loss": 0.5752276182174683, "lr": 3e-05, "grad_norm": 0.35367433659115854, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135906124229, "tokens_per_second": 154919.63366021946, "data_wait_s": 1.2986999990971526e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 928600, "loss": 0.5858592987060547, "lr": 3e-05, "grad_norm": 0.37578295498138514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135920801717, "tokens_per_second": 153644.25579552943, "data_wait_s": 1.4542999906552723e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 928700, "loss": 0.5363717675209045, "lr": 3e-05, "grad_norm": 0.3355081071269844, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135935459183, "tokens_per_second": 153776.51483855143, "data_wait_s": 6.9359998633444775e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 928800, "loss": 0.5778369307518005, "lr": 3e-05, "grad_norm": 0.3463705909920343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135949889323, "tokens_per_second": 151523.70106611072, "data_wait_s": 1.3439999747788534e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:234cbf13aa1ecd6e486de6fc85850bef602e2760ee3a7830c9ff872d0fb4998f
-size 78531
+oid sha256:39f540c2ad63e5a95731be97f0c5677984bcd8c74939c301e3bbeeef30c6e1f6
+size 80649