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": 74703.80895349976,
-      "required_pages_per_second_upper_estimate": 0.006475136825016552
+      "required_token_positions_per_second": 75938.45905140482,
+      "required_pages_per_second_upper_estimate": 0.006582153166310898
     },
     {
       "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": 9337.97611918747,
-      "required_pages_per_second_upper_estimate": 0.003725058289128558
+      "required_token_positions_per_second": 9492.307381425602,
+      "required_pages_per_second_upper_estimate": 0.003786623337093347
     },
     {
       "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": 18675.95223837494,
-      "required_pages_per_second_upper_estimate": 0.004320352864814652
+      "required_token_positions_per_second": 18984.614762851204,
+      "required_pages_per_second_upper_estimate": 0.00439175650757734
     },
     {
       "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": 28013.92835756241,
-      "required_pages_per_second_upper_estimate": 0.0017059012592524957
+      "required_token_positions_per_second": 28476.92214427681,
+      "required_pages_per_second_upper_estimate": 0.0017340951517227376
     },
     {
       "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": 28013.92835756241,
-      "required_pages_per_second_upper_estimate": 0.00215658975248211
+      "required_token_positions_per_second": 28476.92214427681,
+      "required_pages_per_second_upper_estimate": 0.002192232295832215
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28013.92835756241,
-      "required_pages_per_second_upper_estimate": 0.34783923361389685
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28476.92214427681,
+      "required_pages_per_second_upper_estimate": 0.35358806690436456
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2943498913515441,
-    "producer_p95_seconds": 0.22377110199886374,
-    "gpu_update_seconds": 0.9869376278509819,
-    "producer_batches_per_second": 3.3973173742595106,
-    "gpu_batches_per_second": 1.0132352559882238,
-    "producer_headroom": 3.3529403504086077,
+    "producer_batch_seconds": 0.08756026350775414,
+    "producer_p95_seconds": 0.22659823800131562,
+    "gpu_update_seconds": 0.9708914418462389,
+    "producer_batches_per_second": 11.42070569387268,
+    "gpu_batches_per_second": 1.0299812696859378,
+    "producer_headroom": 11.088265418025596,
     "sustainable": true,
-    "buffer_seconds": 60.2031952989099,
+    "buffer_seconds": 60.19526939446681,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 186614.23478917993,
+    "generation_bytes_per_second": 186559.715878379,
     "upload_bytes_per_second": 6298586.757238512,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 40.043070628705486,
+    "estimated_full_shards_per_hour": 40.0313721395829,
     "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

@@ -171,3 +171,4 @@
 {"step": 17100, "loss": 0.6575552709400654, "lr": 0.0003, "grad_norm": 0.1446292805032547, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2502147483, "tokens_per_second": 136587.49712598158, "data_wait_s": 1.0157000360777602e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 17200, "loss": 0.6627637622877955, "lr": 0.0003, "grad_norm": 0.14520324430035478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2516775760, "tokens_per_second": 149344.46905586615, "data_wait_s": 1.0747997293947265e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 17300, "loss": 0.6741339080035686, "lr": 0.0003, "grad_norm": 0.13986137077484798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2531337238, "tokens_per_second": 147541.16995837618, "data_wait_s": 9.084000339498743e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 17400, "loss": 0.6582036688923836, "lr": 0.0003, "grad_norm": 0.14631516266041225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2546013116, "tokens_per_second": 150291.4679919444, "data_wait_s": 1.1026000720448792e-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 sha256:6e2a74422afad66c74a70c39d8efc567e3ea1c056a088c1393513a147a0bf7d2
-size 363154
+oid sha256:3d81a7b09dc099a8f46bcdec4747fbb1fc51d1d864c652961408ec256898bc44
+size 365272