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": 75792.54215271515,
-      "required_pages_per_second_upper_estimate": 0.006569505459355476
+      "required_token_positions_per_second": 75842.57462698415,
+      "required_pages_per_second_upper_estimate": 0.006573842147418966
     },
     {
       "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": 9474.067769089394,
-      "required_pages_per_second_upper_estimate": 0.003779347283025927
+      "required_token_positions_per_second": 9480.321828373018,
+      "required_pages_per_second_upper_estimate": 0.0037818421207806837
     },
     {
       "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": 18948.135538178787,
-      "required_pages_per_second_upper_estimate": 0.004383317680961803
+      "required_token_positions_per_second": 18960.643656746037,
+      "required_pages_per_second_upper_estimate": 0.0043862112140569684
     },
     {
       "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": 28422.203307268177,
-      "required_pages_per_second_upper_estimate": 0.0017307630616364622
+      "required_token_positions_per_second": 28440.965485119053,
+      "required_pages_per_second_upper_estimate": 0.0017319055798300411
     },
     {
       "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": 28422.203307268177,
-      "required_pages_per_second_upper_estimate": 0.0021880198882878554
+      "required_token_positions_per_second": 28440.965485119053,
+      "required_pages_per_second_upper_estimate": 0.002189464252675848
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28422.203307268177,
-      "required_pages_per_second_upper_estimate": 0.3529086420808843
+      "required_token_positions_per_second": 28440.965485119053,
+      "required_pages_per_second_upper_estimate": 0.3531416055368222
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.060435667835953666,
-    "producer_p95_seconds": 0.18357561800075928,
-    "gpu_update_seconds": 0.9727606160965643,
-    "producer_batches_per_second": 16.54652022236928,
-    "gpu_batches_per_second": 1.0280021450834844,
-    "producer_headroom": 16.095803205766202,
+    "producer_batch_seconds": 0.080577971249987,
+    "producer_p95_seconds": 0.18257642799926543,
+    "gpu_update_seconds": 0.9721188971051651,
+    "producer_batches_per_second": 12.410339755236285,
+    "gpu_batches_per_second": 1.028680753946725,
+    "producer_headroom": 12.06432579556068,
     "sustainable": true,
-    "buffer_seconds": 60.311158197986984,
+    "buffer_seconds": 60.271371620520235,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 160,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354995.1417486595,
-    "upload_bytes_per_second": 33601.52605635909,
+    "generation_bytes_per_second": 905156.4916824569,
+    "upload_bytes_per_second": 607793.4295288165,
     "upload_sustainable": false,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.17369355530585,
+    "estimated_full_shards_per_hour": 97.11275607516899,
     "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": [
@@ -184,6 +184,12 @@
       "bytes": 27860,
       "pages": 1,
       "shards": 1
+    },
+    {
+      "seconds": 0.7566531879992908,
+      "bytes": 935968,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790696019647648547-3fdea0a8/train.jsonl

@@ -47,3 +47,4 @@
 {"step": 279700, "loss": 0.6038200240582228, "lr": 0.0003, "grad_norm": 0.14256179918893477, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40940438122, "tokens_per_second": 151937.09150502417, "data_wait_s": 7.476999599020928e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 279800, "loss": 0.631115909665823, "lr": 0.0003, "grad_norm": 0.14944488747335424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40955163321, "tokens_per_second": 150272.22707045428, "data_wait_s": 1.2073000107193366e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 279900, "loss": 0.5793531276285648, "lr": 0.0003, "grad_norm": 0.14385244832193944, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40969928070, "tokens_per_second": 150022.49651742345, "data_wait_s": 1.4277999980549794e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 280000, "loss": 0.5934325978159904, "lr": 0.0003, "grad_norm": 0.18046638129481082, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40984455791, "tokens_per_second": 148115.39999635564, "data_wait_s": 1.1576999895623885e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790696019.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.22407.0.1790696019647648547-3fdea0a8

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ad7cd4e83973437e4783f47d5b27ab2d168f014efa6245f4ce2da531141d210d
-size 103947
+oid sha256:ed2c20c6353f7fad42e447a86ccaf8ca69ee5b99e83e90ef4e7818bfaa85954c
+size 106065