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": 76179.23621280523,
-      "required_pages_per_second_upper_estimate": 0.00527025632138268
+      "required_token_positions_per_second": 76522.55523776438,
+      "required_pages_per_second_upper_estimate": 0.005294007928139243
     },
     {
       "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": 9522.404526600654,
-      "required_pages_per_second_upper_estimate": 0.003986831924869144
+      "required_token_positions_per_second": 9565.319404720547,
+      "required_pages_per_second_upper_estimate": 0.004004799488173381
     },
     {
       "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": 19044.80905320131,
-      "required_pages_per_second_upper_estimate": 0.00396653148039256
+      "required_token_positions_per_second": 19130.638809441094,
+      "required_pages_per_second_upper_estimate": 0.0039844075551344225
     },
     {
       "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": 28567.213579801963,
-      "required_pages_per_second_upper_estimate": 0.0017125608344996485
+      "required_token_positions_per_second": 28695.95821416164,
+      "required_pages_per_second_upper_estimate": 0.001720278878747837
     },
     {
       "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": 28567.213579801963,
-      "required_pages_per_second_upper_estimate": 0.0022209752019365777
+      "required_token_positions_per_second": 28695.95821416164,
+      "required_pages_per_second_upper_estimate": 0.0022309845309702436
     },
     {
       "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": 28567.213579801963,
-      "required_pages_per_second_upper_estimate": 0.3664295427175378
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28695.95821416164,
+      "required_pages_per_second_upper_estimate": 0.36808094065188546
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09312377078930467,
-    "producer_p95_seconds": 0.33125174699671334,
-    "gpu_update_seconds": 0.9678227777716523,
-    "producer_batches_per_second": 10.738396775862203,
-    "gpu_batches_per_second": 1.0332470189453835,
-    "producer_headroom": 10.392864996429111,
+    "producer_batch_seconds": 0.11274654535122863,
+    "producer_p95_seconds": 0.3352911670008325,
+    "gpu_update_seconds": 0.9634806335324092,
+    "producer_batches_per_second": 8.86945135999329,
+    "gpu_batches_per_second": 1.0379035812413788,
+    "producer_headroom": 8.545544615411224,
     "sustainable": true,
-    "buffer_seconds": 60.00501222184244,
+    "buffer_seconds": 60.69927991254178,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 423108.64731044765,
-    "upload_bytes_per_second": 5009924.907891594,
+    "generation_bytes_per_second": 440368.8855646504,
+    "upload_bytes_per_second": 4969501.956989679,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.78926624760696,
+    "estimated_full_shards_per_hour": 94.49291157917628,
     "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": [
-    {
-      "seconds": 1.2126570350010297,
-      "bytes": 6557056,
-      "pages": 45,
-      "shards": 2
-    },
     {
       "seconds": 0.8083091259977664,
       "bytes": 741591,
@@ -562,6 +556,12 @@
       "bytes": 7325771,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 0.8256457960014814,
+      "bytes": 1213741,
+      "pages": 36,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -381,3 +381,4 @@
 {"step": 478300, "loss": 0.5612347722053528, "lr": 0.0003, "grad_norm": 0.1444592162545219, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70007921025, "tokens_per_second": 149904.0171321939, "data_wait_s": 1.1409996659494936e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 478400, "loss": 0.5991029143333435, "lr": 0.0003, "grad_norm": 0.1899080482951865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70022483508, "tokens_per_second": 152213.58236681233, "data_wait_s": 1.0787996870931238e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 478500, "loss": 0.560171365737915, "lr": 0.0003, "grad_norm": 0.16747739114457763, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70037184439, "tokens_per_second": 152712.71266018532, "data_wait_s": 9.809999028220773e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 478600, "loss": 0.5615736842155457, "lr": 0.0003, "grad_norm": 0.1617892725936926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70051902110, "tokens_per_second": 152980.99716700756, "data_wait_s": 1.5670004358980805e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:881f0247a4f54887ca7934d0c41d234a49c35a809f980c6c4a4fa784436957df
-size 816840
+oid sha256:b278af514b3a91c8fc9fa1a3a4c35896f7599aba863c51337dff4561a6b9bc5f
+size 818958