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": 75898.14008863931,
-      "required_pages_per_second_upper_estimate": 0.005250809439280013
+      "required_token_positions_per_second": 76548.08554960876,
+      "required_pages_per_second_upper_estimate": 0.005295774174351123
     },
     {
       "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": 9487.267511079914,
-      "required_pages_per_second_upper_estimate": 0.003972120790214931
+      "required_token_positions_per_second": 9568.510693701095,
+      "required_pages_per_second_upper_estimate": 0.004006135614227845
     },
     {
       "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": 18974.535022159827,
-      "required_pages_per_second_upper_estimate": 0.003951895252977444
+      "required_token_positions_per_second": 19137.02138740219,
+      "required_pages_per_second_upper_estimate": 0.003985736877803819
     },
     {
       "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": 28461.80253323974,
-      "required_pages_per_second_upper_estimate": 0.001706241603211592
+      "required_token_positions_per_second": 28705.532081103283,
+      "required_pages_per_second_upper_estimate": 0.0017208528174525403
     },
     {
       "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": 28461.80253323974,
-      "required_pages_per_second_upper_estimate": 0.0022127799567205585
+      "required_token_positions_per_second": 28705.532081103283,
+      "required_pages_per_second_upper_estimate": 0.002231728857013966
     },
     {
       "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": 28461.80253323974,
-      "required_pages_per_second_upper_estimate": 0.3650774429938013
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28705.532081103283,
+      "required_pages_per_second_upper_estimate": 0.3682037439373954
     }
   ],
   "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": 0.7176326357421772,
-    "producer_p95_seconds": 0.22813570799917215,
-    "gpu_update_seconds": 0.9714072033108471,
-    "producer_batches_per_second": 1.3934706285560687,
-    "gpu_batches_per_second": 1.029434408754331,
-    "producer_headroom": 1.353627406181459,
+    "producer_batch_seconds": 0.08237858583609636,
+    "producer_p95_seconds": 0.2233849910007848,
+    "gpu_update_seconds": 0.9631592935426042,
+    "producer_batches_per_second": 12.139077041084912,
+    "gpu_batches_per_second": 1.0382498582574973,
+    "producer_headroom": 11.691864867150588,
     "sustainable": true,
-    "buffer_seconds": 60.22724660527252,
+    "buffer_seconds": 60.67903549318407,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297630.92124157696,
-    "upload_bytes_per_second": 4448315.030617855,
+    "generation_bytes_per_second": 285198.63712742296,
+    "upload_bytes_per_second": 4539089.648220684,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.86466720519525,
+    "estimated_full_shards_per_hour": 61.19698844306008,
     "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": 1.1898623689994565,
-      "bytes": 6904260,
-      "pages": 25,
-      "shards": 3
-    },
     {
       "seconds": 1.1612172679997457,
       "bytes": 6026166,
@@ -562,6 +556,12 @@
       "bytes": 858247,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.4054743929991673,
+      "bytes": 16596361,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -190,3 +190,4 @@
 {"step": 459200, "loss": 0.5952470898628235, "lr": 0.0003, "grad_norm": 0.1535267059201879, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67212758806, "tokens_per_second": 150534.14906293093, "data_wait_s": 1.428700124961324e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 459300, "loss": 0.5798925757408142, "lr": 0.0003, "grad_norm": 0.1686032957160167, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67227281839, "tokens_per_second": 151925.63496468656, "data_wait_s": 8.195998816518113e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 459400, "loss": 0.5622214078903198, "lr": 0.0003, "grad_norm": 0.16292578632419594, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67241918117, "tokens_per_second": 122496.68243172229, "data_wait_s": 1.2632001016754657e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 459500, "loss": 0.591704785823822, "lr": 0.0003, "grad_norm": 0.18903362584495223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67256612063, "tokens_per_second": 152532.32364607448, "data_wait_s": 1.0562998795649037e-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:d383132a640a65b859a470226820bc60ea6944b9e467898b47e42c8751dca68b
-size 409170
+oid sha256:5fbc429a6cead5f2e043d2d87cfd4c69b9a01593b08a54bd33a4b5827b7a4ba7
+size 411288