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": 76619.00236794731,
-      "required_pages_per_second_upper_estimate": 0.005300680364393472
+      "required_token_positions_per_second": 77443.0087133015,
+      "required_pages_per_second_upper_estimate": 0.005357687035323213
     },
     {
       "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": 9577.375295993414,
-      "required_pages_per_second_upper_estimate": 0.004009847038093693
+      "required_token_positions_per_second": 9680.376089162688,
+      "required_pages_per_second_upper_estimate": 0.004052971319292519
     },
     {
       "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": 19154.750591986827,
-      "required_pages_per_second_upper_estimate": 0.003989429403568233
+      "required_token_positions_per_second": 19360.752178325376,
+      "required_pages_per_second_upper_estimate": 0.004032334101375393
     },
     {
       "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": 28732.125887980237,
-      "required_pages_per_second_upper_estimate": 0.0017224470755684217
+      "required_token_positions_per_second": 29041.128267488064,
+      "required_pages_per_second_upper_estimate": 0.0017409712964006016
     },
     {
       "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": 28732.125887980237,
-      "required_pages_per_second_upper_estimate": 0.0022337964085248547
+      "required_token_positions_per_second": 29041.128267488064,
+      "required_pages_per_second_upper_estimate": 0.002257819984373761
     },
     {
       "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": 28732.125887980237,
-      "required_pages_per_second_upper_estimate": 0.36854486073780784
+      "required_token_positions_per_second": 29041.128267488064,
+      "required_pages_per_second_upper_estimate": 0.3725084114812286
     }
   ],
   "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": 0.22303313259372004,
-    "producer_p95_seconds": 0.15788651699949696,
-    "gpu_update_seconds": 0.9622678150511038,
-    "producer_batches_per_second": 4.483638768691881,
-    "gpu_batches_per_second": 1.039211729165952,
-    "producer_headroom": 4.314461281427558,
+    "producer_batch_seconds": 0.0956876262734383,
+    "producer_p95_seconds": 0.16289937200053828,
+    "gpu_update_seconds": 0.9520291272895313,
+    "producer_batches_per_second": 10.450672035090369,
+    "gpu_batches_per_second": 1.0503880305081041,
+    "producer_headroom": 9.949344177156194,
     "sustainable": true,
-    "buffer_seconds": 60.62287234821954,
+    "buffer_seconds": 60.92986414653,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 474469.8298379507,
-    "upload_bytes_per_second": 5085897.5587612,
+    "generation_bytes_per_second": 467019.99733200076,
+    "upload_bytes_per_second": 5368592.714147615,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.81018039087192,
+    "estimated_full_shards_per_hour": 100.21161975832001,
     "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": [
@@ -208,6 +208,12 @@
       "bytes": 11539917,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.530777683999986,
+      "bytes": 10014604,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -41,3 +41,4 @@
 {"step": 764100, "loss": 0.5349744558334351, "lr": 3e-05, "grad_norm": 0.23846536295197823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111841718650, "tokens_per_second": 152247.7435756337, "data_wait_s": 1.2776999938068911e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 764200, "loss": 0.5094667673110962, "lr": 3e-05, "grad_norm": 0.2447531296317043, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111856435679, "tokens_per_second": 153338.01638305493, "data_wait_s": 1.0163000297325198e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 764300, "loss": 0.536858320236206, "lr": 3e-05, "grad_norm": 0.23199676786754267, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111871174152, "tokens_per_second": 153852.91221550235, "data_wait_s": 1.0695999662857503e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 764400, "loss": 0.5274600982666016, "lr": 3e-05, "grad_norm": 0.3359093690390421, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111885697494, "tokens_per_second": 151168.8128846808, "data_wait_s": 9.989999853132758e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b3266e2da0e6551950c169fba44796af2d199ce88a87db186ee3e110b5a3d06b
-size 91239
+oid sha256:87ffbdcfe31dbbe84a18bcfa087384248ef66673eef4c0426be51f2f567fd757
+size 93357