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": 76823.68449603615,
-      "required_pages_per_second_upper_estimate": 0.005314840748942628
+      "required_token_positions_per_second": 77606.28866600104,
+      "required_pages_per_second_upper_estimate": 0.005368983121312654
     },
     {
       "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": 9602.960562004519,
-      "required_pages_per_second_upper_estimate": 0.004020559054691433
+      "required_token_positions_per_second": 9700.78608325013,
+      "required_pages_per_second_upper_estimate": 0.004061516557607789
     },
     {
       "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": 19205.921124009037,
-      "required_pages_per_second_upper_estimate": 0.004000086875930783
+      "required_token_positions_per_second": 19401.57216650026,
+      "required_pages_per_second_upper_estimate": 0.004040835828416741
     },
     {
       "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": 28808.88168601356,
-      "required_pages_per_second_upper_estimate": 0.0017270484684612018
+      "required_token_positions_per_second": 29102.35824975039,
+      "required_pages_per_second_upper_estimate": 0.0017446419403444585
     },
     {
       "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": 28808.88168601356,
-      "required_pages_per_second_upper_estimate": 0.00223976383420887
+      "required_token_positions_per_second": 29102.35824975039,
+      "required_pages_per_second_upper_estimate": 0.0022625803461724277
     },
     {
       "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": 28808.88168601356,
-      "required_pages_per_second_upper_estimate": 0.36952940170102433
+      "required_token_positions_per_second": 29102.35824975039,
+      "required_pages_per_second_upper_estimate": 0.37329380395005696
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05514860415624412,
-    "producer_p95_seconds": 0.18136322800000926,
-    "gpu_update_seconds": 0.9597040350727271,
-    "producer_batches_per_second": 18.132825214702674,
-    "gpu_batches_per_second": 1.0419879082036154,
-    "producer_headroom": 17.402145525818643,
+    "producer_batch_seconds": 0.1818507274687562,
+    "producer_p95_seconds": 0.18118723999941722,
+    "gpu_update_seconds": 0.9500261031333135,
+    "producer_batches_per_second": 5.499015670266209,
+    "gpu_batches_per_second": 1.0526026566026616,
+    "producer_headroom": 5.224208428292032,
     "sustainable": true,
-    "buffer_seconds": 60.461354209581806,
+    "buffer_seconds": 60.80167060053206,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 555741.9408485374,
+    "generation_bytes_per_second": 537992.772335077,
     "upload_bytes_per_second": 3096600.82148347,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 119.24928349582758,
+    "estimated_full_shards_per_hour": 115.44072511233553,
     "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-v1/1791143623306220418-eb05a537/train.jsonl

@@ -76,3 +76,4 @@
 {"step": 692600, "loss": 0.5743834972381592, "lr": 8e-05, "grad_norm": 0.19790142158113494, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101377989277, "tokens_per_second": 152639.34258261672, "data_wait_s": 1.735400019242661e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 692700, "loss": 0.5638164281845093, "lr": 8e-05, "grad_norm": 0.21678316106732431, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101392716786, "tokens_per_second": 153700.706609766, "data_wait_s": 1.4446999557549134e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 692800, "loss": 0.5742419362068176, "lr": 8e-05, "grad_norm": 0.21497819229656606, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101407438831, "tokens_per_second": 154094.2638992538, "data_wait_s": 1.4145999557513278e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 692900, "loss": 0.5348964929580688, "lr": 8e-05, "grad_norm": 0.20250001247449992, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101422163664, "tokens_per_second": 153221.58892017504, "data_wait_s": 1.732199962134473e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:37d0e5105da574e9bd0fe7215e33d6044d20702e8ba820b66abc7c65c0976ab0
-size 166152
+oid sha256:2e24380a1640349ef29ea4b20cba47bb95919d9ffba7ad7b429383fc39067fd4
+size 168270