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": 77383.66832653261,
-      "required_pages_per_second_upper_estimate": 0.005353581729677874
+      "required_token_positions_per_second": 77097.62481024813,
+      "required_pages_per_second_upper_estimate": 0.005333792575508914
     },
     {
       "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": 9672.958540816577,
-      "required_pages_per_second_upper_estimate": 0.004049865746696026
+      "required_token_positions_per_second": 9637.203101281017,
+      "required_pages_per_second_upper_estimate": 0.00403489569082097
     },
     {
       "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": 19345.917081633153,
-      "required_pages_per_second_upper_estimate": 0.0040292443419622116
+      "required_token_positions_per_second": 19274.406202562033,
+      "required_pages_per_second_upper_estimate": 0.004014350511720397
     },
     {
       "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": 29018.875622449734,
-      "required_pages_per_second_upper_estimate": 0.0017396372843083753
+      "required_token_positions_per_second": 28911.609303843048,
+      "required_pages_per_second_upper_estimate": 0.0017332068322940895
     },
     {
       "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": 29018.875622449734,
-      "required_pages_per_second_upper_estimate": 0.0022560899391010723
+      "required_token_positions_per_second": 28911.609303843048,
+      "required_pages_per_second_upper_estimate": 0.002247750454644075
     },
     {
       "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": 29018.875622449734,
-      "required_pages_per_second_upper_estimate": 0.3722229784437056
+      "required_token_positions_per_second": 28911.609303843048,
+      "required_pages_per_second_upper_estimate": 0.37084708128221866
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09248464625005681,
-    "producer_p95_seconds": 0.22130654600005073,
-    "gpu_update_seconds": 0.9527591750870875,
-    "producer_batches_per_second": 10.812605557210375,
-    "gpu_batches_per_second": 1.0495831750017988,
-    "producer_headroom": 10.301809151229815,
+    "producer_batch_seconds": 0.07953494893752122,
+    "producer_p95_seconds": 0.2104277380003623,
+    "gpu_update_seconds": 0.9562940516190814,
+    "producer_batches_per_second": 12.573089105589936,
+    "gpu_batches_per_second": 1.045703461510527,
+    "producer_headroom": 12.023570322152333,
     "sustainable": true,
-    "buffer_seconds": 60.02382803048651,
+    "buffer_seconds": 60.24652525200213,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 366315.7184747449,
-    "upload_bytes_per_second": 3555161.2142389095,
+    "generation_bytes_per_second": 386742.6981299991,
+    "upload_bytes_per_second": 3652605.4285908067,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.602825791185,
+    "estimated_full_shards_per_hour": 82.98598010945301,
     "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": 1.9101255940004194,
-      "bytes": 862278,
-      "pages": 25,
-      "shards": 1
-    },
     {
       "seconds": 3.149820212999657,
       "bytes": 17435942,
@@ -562,6 +556,12 @@
       "bytes": 7535662,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.3609303779994661,
+      "bytes": 10666543,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -157,3 +157,4 @@
 {"step": 940700, "loss": 0.5371867418289185, "lr": 3e-05, "grad_norm": 0.34323661647359505, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137691947860, "tokens_per_second": 154600.14354784923, "data_wait_s": 1.632999919820577e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 940800, "loss": 0.5152949690818787, "lr": 3e-05, "grad_norm": 0.3460394793445464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137706454781, "tokens_per_second": 151085.31234444628, "data_wait_s": 1.1926000297535211e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 940900, "loss": 0.5127492547035217, "lr": 3e-05, "grad_norm": 0.33479694834108586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137720947183, "tokens_per_second": 150850.5953767437, "data_wait_s": 1.6580999727011658e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 941000, "loss": 0.5244556069374084, "lr": 3e-05, "grad_norm": 0.336522169660098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137735627709, "tokens_per_second": 153399.61329272506, "data_wait_s": 1.2283999240025878e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:776e4def61115e32e6f5cf4bcf0f5a381ede19526b60407719f59b31b947ecfb
-size 339276
+oid sha256:682f75aef802a909d080fdd2c5873072770ebb89ea06504eb202145780bc29de
+size 341394