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": 76549.64468686633,
-      "required_pages_per_second_upper_estimate": 0.005295882039084298
+      "required_token_positions_per_second": 76798.44742589604,
+      "required_pages_per_second_upper_estimate": 0.005313094789872265
     },
     {
       "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": 9568.705585858292,
-      "required_pages_per_second_upper_estimate": 0.004006217211504252
+      "required_token_positions_per_second": 9599.805928237005,
+      "required_pages_per_second_upper_estimate": 0.004019238275409219
     },
     {
       "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": 19137.411171716583,
-      "required_pages_per_second_upper_estimate": 0.003985818059597203
+      "required_token_positions_per_second": 19199.61185647401,
+      "required_pages_per_second_upper_estimate": 0.00399877282188981
     },
     {
       "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": 28706.116757574873,
-      "required_pages_per_second_upper_estimate": 0.0017208878679142627
+      "required_token_positions_per_second": 28799.41778471102,
+      "required_pages_per_second_upper_estimate": 0.001726481122028655
     },
     {
       "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": 28706.116757574873,
-      "required_pages_per_second_upper_estimate": 0.002231774313037912
+      "required_token_positions_per_second": 28799.41778471102,
+      "required_pages_per_second_upper_estimate": 0.0022390280575099
     },
     {
       "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": 28706.116757574873,
-      "required_pages_per_second_upper_estimate": 0.36821124353939627
+      "required_token_positions_per_second": 28799.41778471102,
+      "required_pages_per_second_upper_estimate": 0.36940800893666087
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.23830814819531554,
-    "producer_p95_seconds": 0.24824349000118673,
-    "gpu_update_seconds": 0.9631396762400591,
-    "producer_batches_per_second": 4.196247621295801,
-    "gpu_batches_per_second": 1.0382710054099709,
-    "producer_headroom": 4.0415725753979554,
+    "producer_batch_seconds": 0.11187522788289073,
+    "producer_p95_seconds": 0.24297064900019905,
+    "gpu_update_seconds": 0.9600194075686391,
+    "producer_batches_per_second": 8.938529278767456,
+    "gpu_batches_per_second": 1.0416456085326613,
+    "producer_headroom": 8.581161582737268,
     "sustainable": true,
-    "buffer_seconds": 60.677799603123724,
+    "buffer_seconds": 60.48122267682427,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 498435.05610526673,
-    "upload_bytes_per_second": 3960080.2558228336,
+    "generation_bytes_per_second": 518336.7678302552,
+    "upload_bytes_per_second": 3968843.4900915665,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.95256006592275,
+    "estimated_full_shards_per_hour": 111.22300411396735,
     "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": 1.5363667999999961,
-      "bytes": 6120972,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.693875957000273,
       "bytes": 10707952,
@@ -562,6 +556,12 @@
       "bytes": 5842602,
       "pages": 41,
       "shards": 2
+    },
+    {
+      "seconds": 1.6693870859999151,
+      "bytes": 7612861,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -138,3 +138,4 @@
 {"step": 813800, "loss": 0.5414049029350281, "lr": 3e-05, "grad_norm": 0.2815565741459654, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119116976455, "tokens_per_second": 152972.4789200797, "data_wait_s": 1.6899000911507756e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 813900, "loss": 0.529843270778656, "lr": 3e-05, "grad_norm": 0.2860852396890787, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119131526765, "tokens_per_second": 150183.54668206448, "data_wait_s": 1.9241000700276345e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 814000, "loss": 0.5257862210273743, "lr": 3e-05, "grad_norm": 0.2666706961559004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119146200406, "tokens_per_second": 152149.18129974138, "data_wait_s": 8.697999874129891e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 814100, "loss": 0.5360915660858154, "lr": 3e-05, "grad_norm": 0.27407967578988174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119160974846, "tokens_per_second": 152299.70054497197, "data_wait_s": 1.3650000255438499e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:57918a817f0657d01b8ffc17d7d000c16652ecbdbde373e402c60ffd647aad15
-size 298251
+oid sha256:58b78e522e2993c08b34e34b1c345b9c2e656caaae001ef532773d2bdf49315f
+size 300369