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": 75750.32847717725,
-      "required_pages_per_second_upper_estimate": 0.005240583489028877
+      "required_token_positions_per_second": 77304.04976607049,
+      "required_pages_per_second_upper_estimate": 0.005348073532924592
     },
     {
       "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": 9468.791059647156,
-      "required_pages_per_second_upper_estimate": 0.003964385085832215
+      "required_token_positions_per_second": 9663.00622075881,
+      "required_pages_per_second_upper_estimate": 0.004045698918115915
     },
     {
       "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": 18937.58211929431,
-      "required_pages_per_second_upper_estimate": 0.003944198937824139
+      "required_token_positions_per_second": 19326.01244151762,
+      "required_pages_per_second_upper_estimate": 0.004025098730346805
     },
     {
       "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": 28406.37317894147,
-      "required_pages_per_second_upper_estimate": 0.0017029186980571341
+      "required_token_positions_per_second": 28989.018662276434,
+      "required_pages_per_second_upper_estimate": 0.0017378474051349246
     },
     {
       "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": 28406.37317894147,
-      "required_pages_per_second_upper_estimate": 0.002208470568231831
+      "required_token_positions_per_second": 28989.018662276434,
+      "required_pages_per_second_upper_estimate": 0.002253768691774487
     },
     {
       "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": 28406.37317894147,
-      "required_pages_per_second_upper_estimate": 0.36436645475226676
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28989.018662276434,
+      "required_pages_per_second_upper_estimate": 0.3718400054165087
     }
   ],
   "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": 0.09791073607783574,
-    "producer_p95_seconds": 0.33399655000539497,
-    "gpu_update_seconds": 0.9733027101290187,
-    "producer_batches_per_second": 10.213384558819307,
-    "gpu_batches_per_second": 1.0274295854651863,
-    "producer_headroom": 9.940714870688703,
+    "producer_batch_seconds": 0.11561083932855354,
+    "producer_p95_seconds": 0.3393006949991104,
+    "gpu_update_seconds": 0.9537404602101447,
+    "producer_batches_per_second": 8.649707984197812,
+    "gpu_batches_per_second": 1.0485032791621975,
+    "producer_headroom": 8.249576473532185,
     "sustainable": true,
-    "buffer_seconds": 60.34476802799916,
+    "buffer_seconds": 60.08564899323912,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 289790.73978774744,
-    "upload_bytes_per_second": 4902045.104254086,
+    "generation_bytes_per_second": 302427.05894417525,
+    "upload_bytes_per_second": 4838539.670107618,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.18234677528684,
+    "estimated_full_shards_per_hour": 64.89380670780128,
     "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": 1.0987542440016114,
-      "bytes": 12106318,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.2125764899974456,
       "bytes": 7340281,
@@ -562,6 +556,12 @@
       "bytes": 12270542,
       "pages": 42,
       "shards": 4
+    },
+    {
+      "seconds": 1.0424309449954308,
+      "bytes": 6176100,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -384,3 +384,4 @@
 {"step": 595500, "loss": 0.5839543342590332, "lr": 0.0003, "grad_norm": 0.20130340175974581, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87163181146, "tokens_per_second": 153542.6390328098, "data_wait_s": 1.160599640570581e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 595600, "loss": 0.5854461789131165, "lr": 0.0003, "grad_norm": 0.16052127449733206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87177776574, "tokens_per_second": 129848.64415712895, "data_wait_s": 1.2450000212993473e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 595700, "loss": 0.5985738039016724, "lr": 0.0003, "grad_norm": 0.2011437839611576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87192496758, "tokens_per_second": 153321.0131154079, "data_wait_s": 1.0705000022426248e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 595800, "loss": 0.5828372240066528, "lr": 0.0003, "grad_norm": 0.16561907383262034, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87207244523, "tokens_per_second": 153697.60219180965, "data_wait_s": 1.5946003259159625e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bcf5fd9e3c580a41730ae65e080d03bee72ac4bc668d0f14500e301486eed34d
-size 823977
+oid sha256:eff3a2dc55a91fe7d4c0f98bb0c9d806da307c3a2b99a70214af9a01d51f74b7
+size 826095