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": 75274.09139106087,
-      "required_pages_per_second_upper_estimate": 0.006524567461336725
+      "required_token_positions_per_second": 75060.68965776266,
+      "required_pages_per_second_upper_estimate": 0.006506070339956185
     },
     {
       "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": 9409.261423882608,
-      "required_pages_per_second_upper_estimate": 0.003753495062981733
+      "required_token_positions_per_second": 9382.586207220333,
+      "required_pages_per_second_upper_estimate": 0.003742853920225121
     },
     {
       "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": 18818.522847765216,
-      "required_pages_per_second_upper_estimate": 0.004353334066140065
+      "required_token_positions_per_second": 18765.172414440665,
+      "required_pages_per_second_upper_estimate": 0.004340992382325989
     },
     {
       "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": 28227.784271647826,
-      "required_pages_per_second_upper_estimate": 0.00171892396240503
+      "required_token_positions_per_second": 28147.758621661,
+      "required_pages_per_second_upper_estimate": 0.0017140508201829692
     },
     {
       "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": 28227.784271647826,
-      "required_pages_per_second_upper_estimate": 0.0021730529727394645
+      "required_token_positions_per_second": 28147.758621661,
+      "required_pages_per_second_upper_estimate": 0.0021668923766783015
     },
     {
       "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": 28227.784271647826,
-      "required_pages_per_second_upper_estimate": 0.3504946083371348
+      "required_token_positions_per_second": 28147.758621661,
+      "required_pages_per_second_upper_estimate": 0.34950095759292005
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08027426271107174,
-    "producer_p95_seconds": 0.3389405169982638,
-    "gpu_update_seconds": 0.9794605107482642,
-    "producer_batches_per_second": 12.457292863582738,
-    "gpu_batches_per_second": 1.0209702065844846,
-    "producer_headroom": 12.201426430705457,
+    "producer_batch_seconds": 0.3370392857265756,
+    "producer_p95_seconds": 0.3460547989998304,
+    "gpu_update_seconds": 0.9822451716892153,
+    "producer_batches_per_second": 2.967013171310996,
+    "gpu_batches_per_second": 1.0180757603320674,
+    "producer_headroom": 2.9143343618585327,
     "sustainable": true,
-    "buffer_seconds": 60.72655166639238,
+    "buffer_seconds": 60.89920064473135,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 519853.69779412297,
-    "upload_bytes_per_second": 4136662.3159422344,
+    "generation_bytes_per_second": 467861.1129947832,
+    "upload_bytes_per_second": 4077810.5416700984,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.54850197189108,
+    "estimated_full_shards_per_hour": 100.39210359938261,
     "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": 1.011032721002266,
-      "bytes": 10718239,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.0137605279996933,
       "bytes": 6019285,
@@ -562,6 +556,12 @@
       "bytes": 7550162,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.2529126609988452,
+      "bytes": 5669267,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -234,3 +234,4 @@
 {"step": 108400, "loss": 0.5929444441571832, "lr": 0.0003, "grad_norm": 0.15431779799373826, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15862485009, "tokens_per_second": 148121.56615737014, "data_wait_s": 1.4886998542351648e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 108500, "loss": 0.5959181506186724, "lr": 0.0003, "grad_norm": 0.1446121997830395, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15876967855, "tokens_per_second": 146694.06227859564, "data_wait_s": 1.4485998690361157e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 108600, "loss": 0.604346277192235, "lr": 0.0003, "grad_norm": 0.14303170411994448, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15891525165, "tokens_per_second": 147561.17176263384, "data_wait_s": 1.471900031901896e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 108700, "loss": 0.5989245846867561, "lr": 0.0003, "grad_norm": 0.1259789825213902, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15906098309, "tokens_per_second": 147376.48087369718, "data_wait_s": 1.2828997569158673e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:26a5a3ad570b00d01cc497cc6150b8f10d9eccff893b3a80ec0f09083525f42e
-size 503145
+oid sha256:5df21c5d92265631c43fec8ef8f81711aea75820d2f4806d6177bd1712f455d2
+size 505263