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": 76835.47728629693,
-      "required_pages_per_second_upper_estimate": 0.0053156566015359116
+      "required_token_positions_per_second": 76934.58856003432,
+      "required_pages_per_second_upper_estimate": 0.005322513349422903
     },
     {
       "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": 9604.434660787116,
-      "required_pages_per_second_upper_estimate": 0.004021176229069024
+      "required_token_positions_per_second": 9616.82357000429,
+      "required_pages_per_second_upper_estimate": 0.004026363206648411
     },
     {
       "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": 19208.86932157423,
-      "required_pages_per_second_upper_estimate": 0.004000700907734414
+      "required_token_positions_per_second": 19233.64714000858,
+      "required_pages_per_second_upper_estimate": 0.0040058614738792755
     },
     {
       "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": 28813.303982361347,
-      "required_pages_per_second_upper_estimate": 0.001727313578374796
+      "required_token_positions_per_second": 28850.470710012873,
+      "required_pages_per_second_upper_estimate": 0.0017295416669471978
     },
     {
       "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": 28813.303982361347,
-      "required_pages_per_second_upper_estimate": 0.0022401076482982847
+      "required_token_positions_per_second": 28850.470710012873,
+      "required_pages_per_second_upper_estimate": 0.002242997197894033
     },
     {
       "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": 28813.303982361347,
-      "required_pages_per_second_upper_estimate": 0.3695861261702819
+      "required_token_positions_per_second": 28850.470710012873,
+      "required_pages_per_second_upper_estimate": 0.37006286104607267
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10786858498443053,
-    "producer_p95_seconds": 0.2192892650000431,
-    "gpu_update_seconds": 0.9595567386831197,
-    "producer_batches_per_second": 9.27053970481153,
-    "gpu_batches_per_second": 1.0421478581583241,
-    "producer_headroom": 8.895608844981323,
+    "producer_batch_seconds": 0.06315037115626865,
+    "producer_p95_seconds": 0.20052172400028212,
+    "gpu_update_seconds": 0.9583205860972126,
+    "producer_batches_per_second": 15.83521967789313,
+    "gpu_batches_per_second": 1.0434921408424793,
+    "producer_headroom": 15.175217002696659,
     "sustainable": true,
-    "buffer_seconds": 60.452074537036545,
+    "buffer_seconds": 60.37419692412439,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 299834.259731405,
-    "upload_bytes_per_second": 3924838.1980689494,
+    "generation_bytes_per_second": 336524.7582239766,
+    "upload_bytes_per_second": 3934582.915120257,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.33745235401737,
+    "estimated_full_shards_per_hour": 72.21037921943162,
     "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": [
@@ -550,6 +550,12 @@
       "bytes": 9820718,
       "pages": 47,
       "shards": 3
+    },
+    {
+      "seconds": 1.6360844549999456,
+      "bytes": 7454978,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -116,3 +116,4 @@
 {"step": 811600, "loss": 0.521442174911499, "lr": 3e-05, "grad_norm": 0.27802499530849445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118795037952, "tokens_per_second": 149513.28307379616, "data_wait_s": 1.6496000171173364e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 811700, "loss": 0.5145800113677979, "lr": 3e-05, "grad_norm": 0.27895607161056146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118809508071, "tokens_per_second": 149904.37313916325, "data_wait_s": 1.5009998605819419e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 811800, "loss": 0.5481447577476501, "lr": 3e-05, "grad_norm": 0.2835895224033298, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118824178708, "tokens_per_second": 152101.93591095626, "data_wait_s": 1.542999962111935e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 811900, "loss": 0.5262207984924316, "lr": 3e-05, "grad_norm": 0.2765897276631927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118838740313, "tokens_per_second": 150479.90507845033, "data_wait_s": 1.8888000340666622e-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:fe6f0b82fed93b507ee6425f96b3f4520c6ca654823a410068a53ac94fb1f544
-size 251655
+oid sha256:7d5a12c729698e8db424ec057988cb3b63da92cf93a6d866e5e4acac3ed6bab5
+size 253773