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": 76182.26382669157,
-      "required_pages_per_second_upper_estimate": 0.005270465778736369
+      "required_token_positions_per_second": 76313.90078282343,
+      "required_pages_per_second_upper_estimate": 0.0052795727288013905
     },
     {
       "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": 9522.782978336447,
-      "required_pages_per_second_upper_estimate": 0.003986990374707949
+      "required_token_positions_per_second": 9539.237597852929,
+      "required_pages_per_second_upper_estimate": 0.003993879580287971
     },
     {
       "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": 19045.565956672894,
-      "required_pages_per_second_upper_estimate": 0.003966689123424804
+      "required_token_positions_per_second": 19078.475195705858,
+      "required_pages_per_second_upper_estimate": 0.003973543250040375
     },
     {
       "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": 28568.348935009337,
-      "required_pages_per_second_upper_estimate": 0.0017126288973107436
+      "required_token_positions_per_second": 28617.71279355879,
+      "required_pages_per_second_upper_estimate": 0.0017155881852565357
     },
     {
       "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": 28568.348935009337,
-      "required_pages_per_second_upper_estimate": 0.0022210634708100512
+      "required_token_positions_per_second": 28617.71279355879,
+      "required_pages_per_second_upper_estimate": 0.0022249012936836044
     },
     {
       "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": 28568.348935009337,
-      "required_pages_per_second_upper_estimate": 0.3664441058350885
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28617.71279355879,
+      "required_pages_per_second_upper_estimate": 0.3670772924097791
     }
   ],
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06709625428906207,
-    "producer_p95_seconds": 0.15089451799940434,
-    "gpu_update_seconds": 0.9677843148337674,
-    "producer_batches_per_second": 14.903961638332744,
-    "gpu_batches_per_second": 1.0332880835868539,
-    "producer_headroom": 14.423820302462607,
+    "producer_batch_seconds": 0.0875303213515295,
+    "producer_p95_seconds": 0.15692202100035502,
+    "gpu_update_seconds": 0.9661149442461017,
+    "producer_batches_per_second": 11.424612460679903,
+    "gpu_batches_per_second": 1.0350735240725835,
+    "producer_headroom": 11.037488830483083,
     "sustainable": true,
-    "buffer_seconds": 60.002627519693576,
+    "buffer_seconds": 60.865241487504406,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 540695.0285090129,
-    "upload_bytes_per_second": 4268269.821667549,
+    "generation_bytes_per_second": 518533.10500759823,
+    "upload_bytes_per_second": 4332706.564006769,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 116.0205663819579,
+    "estimated_full_shards_per_hour": 111.2651335017296,
     "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": [
@@ -358,6 +358,12 @@
       "bytes": 5569114,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 0.9946771380000428,
+      "bytes": 7343286,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -38,3 +38,4 @@
 {"step": 444000, "loss": 0.6055135726928711, "lr": 0.0003, "grad_norm": 0.17247214597601487, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64988420632, "tokens_per_second": 150676.11953123167, "data_wait_s": 1.2222999430377968e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 444100, "loss": 0.5636587142944336, "lr": 0.0003, "grad_norm": 0.1544770824898031, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65002947538, "tokens_per_second": 150526.41377154645, "data_wait_s": 1.225300002261065e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 444200, "loss": 0.5759021639823914, "lr": 0.0003, "grad_norm": 0.17241649940589882, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65017382051, "tokens_per_second": 150116.48436633035, "data_wait_s": 1.4088000170886517e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 444300, "loss": 0.5492525696754456, "lr": 0.0003, "grad_norm": 0.20942501199811075, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65032080794, "tokens_per_second": 153140.43664036572, "data_wait_s": 1.4805000319029205e-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 sha256:4d2ab4cc21832c607fd24e2c0f91a8961802b77716b936b28d0bb25e1a857a61
-size 84885
+oid sha256:5f83a2a5b4ada3b0c9d041ca5d83840b0b15ce3c35cc6f019eac9743ce8de20a
+size 87003