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": 74740.50593037935,
-      "required_pages_per_second_upper_estimate": 0.006478317625964837
+      "required_token_positions_per_second": 74408.8895630938,
+      "required_pages_per_second_upper_estimate": 0.0064495739597225255
     },
     {
       "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": 9342.56324129742,
-      "required_pages_per_second_upper_estimate": 0.003726888160721804
+      "required_token_positions_per_second": 9301.111195386726,
+      "required_pages_per_second_upper_estimate": 0.003710352319844713
     },
     {
       "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": 18685.12648259484,
-      "required_pages_per_second_upper_estimate": 0.004322475164753742
+      "required_token_positions_per_second": 18602.22239077345,
+      "required_pages_per_second_upper_estimate": 0.004303296762173047
     },
     {
       "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": 28027.68972389226,
-      "required_pages_per_second_upper_estimate": 0.0017067392542616743
+      "required_token_positions_per_second": 27903.333586160177,
+      "required_pages_per_second_upper_estimate": 0.0016991666179200214
     },
     {
       "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": 28027.68972389226,
-      "required_pages_per_second_upper_estimate": 0.002157649140556082
+      "required_token_positions_per_second": 27903.333586160177,
+      "required_pages_per_second_upper_estimate": 0.0021480758608336445
     },
     {
       "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": 28027.68972389226,
-      "required_pages_per_second_upper_estimate": 0.3480101037273832
+      "required_token_positions_per_second": 27903.333586160177,
+      "required_pages_per_second_upper_estimate": 0.3464660166899708
     }
   ],
   "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": 0.22550819391401689,
-    "producer_p95_seconds": 0.17854555000030814,
-    "gpu_update_seconds": 0.9864530495509021,
-    "producer_batches_per_second": 4.434428668171969,
-    "gpu_batches_per_second": 1.013732990592168,
-    "producer_headroom": 4.374355682734184,
+    "producer_batch_seconds": 0.06915977300776888,
+    "producer_p95_seconds": 0.16972236600031465,
+    "gpu_update_seconds": 0.9908493519108835,
+    "producer_batches_per_second": 14.459272442777793,
+    "gpu_batches_per_second": 1.0092351557494277,
+    "producer_headroom": 14.326960729029274,
     "sustainable": true,
-    "buffer_seconds": 60.173636022605024,
+    "buffer_seconds": 60.441810466563894,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 540965.9654867498,
-    "upload_bytes_per_second": 5779278.188101049,
+    "generation_bytes_per_second": 524373.5983405693,
+    "upload_bytes_per_second": 5785431.032807307,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 116.0787031502902,
+    "estimated_full_shards_per_hour": 112.51836741125878,
     "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": [
@@ -400,6 +400,12 @@
       "bytes": 5765948,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.0119390079998993,
+      "bytes": 6091488,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -41,3 +41,4 @@
 {"step": 86000, "loss": 0.5379970334470272, "lr": 0.0003, "grad_norm": 0.13144490846414586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12585673405, "tokens_per_second": 148034.74707902598, "data_wait_s": 1.1390000508981757e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 86100, "loss": 0.6486443988978863, "lr": 0.0003, "grad_norm": 0.1270118650594306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12600191174, "tokens_per_second": 146601.8340052045, "data_wait_s": 1.3423000382317696e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 86200, "loss": 0.6315164659172297, "lr": 0.0003, "grad_norm": 0.1346323412636943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12614745751, "tokens_per_second": 147905.91825621613, "data_wait_s": 1.155600057245465e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 86300, "loss": 0.6972997039556503, "lr": 0.0003, "grad_norm": 0.15334566851523973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12629443000, "tokens_per_second": 149320.8702750207, "data_wait_s": 1.1412000276322942e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:731852a61323fcc8657cbe92168dc380c6ca2ac812121f3636488169ec4ead78
-size 92022
+oid sha256:523863e48d9133798ecb0f9734c7be99c2aa68314af9847c4c38b0862b1129be
+size 94140