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": 76171.3299047943,
-      "required_pages_per_second_upper_estimate": 0.005269709344649322
+      "required_token_positions_per_second": 77636.92118801447,
+      "required_pages_per_second_upper_estimate": 0.0053711023502627875
     },
     {
       "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": 9521.416238099288,
-      "required_pages_per_second_upper_estimate": 0.003986418149111432
+      "required_token_positions_per_second": 9704.615148501809,
+      "required_pages_per_second_upper_estimate": 0.004063119707268692
     },
     {
       "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": 19042.832476198575,
-      "required_pages_per_second_upper_estimate": 0.003966119811528732
+      "required_token_positions_per_second": 19409.230297003618,
+      "required_pages_per_second_upper_estimate": 0.004042430815042108
     },
     {
       "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": 28564.24871429786,
-      "required_pages_per_second_upper_estimate": 0.0017123830953397652
+      "required_token_positions_per_second": 29113.84544550543,
+      "required_pages_per_second_upper_estimate": 0.001745330580190041
     },
     {
       "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": 28564.24871429786,
-      "required_pages_per_second_upper_estimate": 0.002220744696684699
+      "required_token_positions_per_second": 29113.84544550543,
+      "required_pages_per_second_upper_estimate": 0.002263473425115548
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28564.24871429786,
-      "required_pages_per_second_upper_estimate": 0.3663915126062757
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 29113.84544550543,
+      "required_pages_per_second_upper_estimate": 0.37344114936321277
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 62,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11683048781264915,
-    "producer_p95_seconds": 0.24077063000004273,
-    "gpu_update_seconds": 0.967923234268744,
-    "producer_batches_per_second": 8.559409608933693,
-    "gpu_batches_per_second": 1.0331397827798705,
-    "producer_headroom": 8.284851432110067,
+    "producer_batch_seconds": 0.09475564381250479,
+    "producer_p95_seconds": 0.2314305459985917,
+    "gpu_update_seconds": 0.9496512596300903,
+    "producer_batches_per_second": 10.553461089649957,
+    "gpu_batches_per_second": 1.0530181367731999,
+    "producer_headroom": 10.022107617243227,
     "sustainable": true,
-    "buffer_seconds": 60.01124052466213,
+    "buffer_seconds": 60.77768061632578,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 397248.9891554783,
+    "generation_bytes_per_second": 397356.10982483625,
     "upload_bytes_per_second": 5384716.570663779,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.24038558958303,
+    "estimated_full_shards_per_hour": 85.26337119158569,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -224,3 +224,4 @@
 {"step": 579500, "loss": 0.6025071740150452, "lr": 0.0003, "grad_norm": 0.16887081732719386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84821567494, "tokens_per_second": 145632.65346693678, "data_wait_s": 1.22540004667826e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 579600, "loss": 0.5668318271636963, "lr": 0.0003, "grad_norm": 0.16806697426411693, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84836347216, "tokens_per_second": 153372.8027465118, "data_wait_s": 8.847000572131947e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 579700, "loss": 0.5969656109809875, "lr": 0.0003, "grad_norm": 0.13797264914020052, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84850919354, "tokens_per_second": 152031.74969802043, "data_wait_s": 9.548999514663592e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 579800, "loss": 0.6223017573356628, "lr": 0.0003, "grad_norm": 0.15240272493459464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84865539341, "tokens_per_second": 152289.9581169443, "data_wait_s": 1.3045999367022887e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:92b81d7e2cf1234541796689a49692705e45df12fe202e2b24b9bbce68d64566
-size 481965
+oid sha256:c23cb191406119af36b90a9b776e8745bbd8ec7b3442b47af91188aa19b209d4
+size 484083