Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 73904.2144922228,
-      "required_pages_per_second_upper_estimate": 0.0064058300036667
+      "required_token_positions_per_second": 74663.59696907108,
+      "required_pages_per_second_upper_estimate": 0.006471651352122617
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9238.02681152785,
-      "required_pages_per_second_upper_estimate": 0.0036851870159278167
+      "required_token_positions_per_second": 9332.949621133885,
+      "required_pages_per_second_upper_estimate": 0.0037230531439021402
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18476.0536230557,
-      "required_pages_per_second_upper_estimate": 0.0042741098382560174
+      "required_token_positions_per_second": 18665.89924226777,
+      "required_pages_per_second_upper_estimate": 0.004318027281091947
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27714.080434583553,
-      "required_pages_per_second_upper_estimate": 0.0016876420939235532
+      "required_token_positions_per_second": 27998.848863401658,
+      "required_pages_per_second_upper_estimate": 0.0017049829971740973
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27714.080434583553,
-      "required_pages_per_second_upper_estimate": 0.002133506629339022
+      "required_token_positions_per_second": 27998.848863401658,
+      "required_pages_per_second_upper_estimate": 0.0021554288912789036
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27714.080434583553,
-      "required_pages_per_second_upper_estimate": 0.3441161259369427
+      "required_token_positions_per_second": 27998.848863401658,
+      "required_pages_per_second_upper_estimate": 0.34765199676424074
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2402350729921423,
-    "producer_p95_seconds": 0.1811735710002722,
-    "gpu_update_seconds": 0.9976156367612655,
-    "producer_batches_per_second": 4.162589531765448,
-    "gpu_batches_per_second": 1.0023900620147408,
-    "producer_headroom": 4.152664406307966,
+    "producer_batch_seconds": 0.049027290085941644,
+    "producer_p95_seconds": 0.16838046599968948,
+    "gpu_update_seconds": 0.9874691682821195,
+    "producer_batches_per_second": 20.396803458789282,
+    "gpu_batches_per_second": 1.0126898460431732,
+    "producer_headroom": 20.14121454706451,
     "sustainable": true,
-    "buffer_seconds": 60.8545538424372,
+    "buffer_seconds": 60.23561926520929,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 415986.3899964062,
-    "upload_bytes_per_second": 3601035.366902879,
+    "generation_bytes_per_second": 416109.04163821554,
+    "upload_bytes_per_second": 3731248.596244409,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.26099562567845,
+    "estimated_full_shards_per_hour": 89.28731381282662,
     "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": [
@@ -364,6 +364,12 @@
       "bytes": 2359076,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.027329577999808,
+      "bytes": 11020380,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -32,3 +32,4 @@
 {"step": 3200, "loss": 0.8346284031867981, "lr": 0.0003, "grad_norm": 0.20888028264026168, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 468054322, "tokens_per_second": 148392.07020325723, "data_wait_s": 1.3474999832396861e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 3300, "loss": 0.8051414489746094, "lr": 0.0003, "grad_norm": 0.25193659655952183, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 482570487, "tokens_per_second": 147370.6245055717, "data_wait_s": 9.698000212665647e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 3400, "loss": 0.8187024667859077, "lr": 0.0003, "grad_norm": 0.2072594781205535, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 497011906, "tokens_per_second": 145624.0575270867, "data_wait_s": 1.6540000160603086e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 3500, "loss": 0.8277444913983345, "lr": 0.0003, "grad_norm": 0.22791912088439592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 511687459, "tokens_per_second": 148971.41919640548, "data_wait_s": 1.034399974741973e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:246663ae23b3901685f34ff340c2ae95a3e5d1ce52d63eebedaf678409b36de7
-size 70948
+oid sha256:d7423ca66f303fa45227a5ddd8b3d1a873c9868218e8122926691f0fd0d11bf1
+size 73031