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": 77279.58539082442,
-      "required_pages_per_second_upper_estimate": 0.005346381030679898
+      "required_token_positions_per_second": 77142.73155474477,
+      "required_pages_per_second_upper_estimate": 0.0053369131647552585
     },
     {
       "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": 9659.948173853052,
-      "required_pages_per_second_upper_estimate": 0.004044418577735755
+      "required_token_positions_per_second": 9642.841444343096,
+      "required_pages_per_second_upper_estimate": 0.0040372563473190706
     },
     {
       "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": 19319.896347706104,
-      "required_pages_per_second_upper_estimate": 0.004023824909298106
+      "required_token_positions_per_second": 19285.68288868619,
+      "required_pages_per_second_upper_estimate": 0.004016699148053848
     },
     {
       "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": 28979.844521559156,
-      "required_pages_per_second_upper_estimate": 0.0017372974294070254
+      "required_token_positions_per_second": 28928.524333029287,
+      "required_pages_per_second_upper_estimate": 0.0017342208624660505
     },
     {
       "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": 28979.844521559156,
-      "required_pages_per_second_upper_estimate": 0.002253055442686499
+      "required_token_positions_per_second": 28928.524333029287,
+      "required_pages_per_second_upper_estimate": 0.0022490655237619546
     },
     {
       "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": 28979.844521559156,
-      "required_pages_per_second_upper_estimate": 0.3717223293898123
+      "required_token_positions_per_second": 28928.524333029287,
+      "required_pages_per_second_upper_estimate": 0.3710640491146764
     }
   ],
   "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": 0.3648400460234029,
-    "producer_p95_seconds": 0.2341208350007946,
-    "gpu_update_seconds": 0.9540423855425328,
-    "producer_batches_per_second": 2.7409271841168836,
-    "gpu_batches_per_second": 1.0481714598364857,
-    "producer_headroom": 2.614960709333249,
+    "producer_batch_seconds": 0.11175185987491432,
+    "producer_p95_seconds": 0.23219462699853466,
+    "gpu_update_seconds": 0.9557348892640458,
+    "producer_batches_per_second": 8.948396931552784,
+    "gpu_batches_per_second": 1.0463152608879227,
+    "producer_headroom": 8.552295150468328,
     "sustainable": true,
-    "buffer_seconds": 60.104670289179566,
+    "buffer_seconds": 60.21129802363489,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384371.1839187751,
-    "upload_bytes_per_second": 5145022.828859301,
+    "generation_bytes_per_second": 384517.8069470008,
+    "upload_bytes_per_second": 5171574.126524703,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.47710836574974,
+    "estimated_full_shards_per_hour": 82.50857025439757,
     "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": 1.1375752950007154,
-      "bytes": 4905532,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.533432411000831,
       "bytes": 7357007,
@@ -562,6 +556,12 @@
       "bytes": 10921992,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.2837640949983324,
+      "bytes": 7864429,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -189,3 +189,4 @@
 {"step": 623300, "loss": 0.5710268616676331, "lr": 0.0001, "grad_norm": 0.1610284059102318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91234490323, "tokens_per_second": 152160.3451231219, "data_wait_s": 1.5248999261530116e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 623400, "loss": 0.5163668990135193, "lr": 0.0001, "grad_norm": 0.15465748857237932, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91248986456, "tokens_per_second": 150213.52096789624, "data_wait_s": 1.3701002899324521e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 623500, "loss": 0.5607584118843079, "lr": 0.0001, "grad_norm": 0.17825670277705963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91263664963, "tokens_per_second": 152793.54941587502, "data_wait_s": 1.3788001524517313e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 623600, "loss": 0.5415359735488892, "lr": 0.0001, "grad_norm": 0.15616119691441457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91278380949, "tokens_per_second": 152902.4984811512, "data_wait_s": 1.572499968460761e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c6b1c34d66ee8885dcc18dfa9fa5196fa1dc320c7505005a1f3812984aeedecf
-size 407835
+oid sha256:671cb54fa92dd39c820b271868a6795e3cfdd6e50e37f8e3660a9ffaf3e71a53
+size 409953