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": 77244.78435101552,
-      "required_pages_per_second_upper_estimate": 0.005343973413996381
+      "required_token_positions_per_second": 77006.65847457536,
+      "required_pages_per_second_upper_estimate": 0.0053274993133360935
     },
     {
       "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": 9655.59804387694,
-      "required_pages_per_second_upper_estimate": 0.0040425972691558
+      "required_token_positions_per_second": 9625.83230932192,
+      "required_pages_per_second_upper_estimate": 0.004030134977676834
     },
     {
       "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": 19311.19608775388,
-      "required_pages_per_second_upper_estimate": 0.004022012874591342
+      "required_token_positions_per_second": 19251.66461864384,
+      "required_pages_per_second_upper_estimate": 0.004009614039526136
     },
     {
       "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": 28966.79413163082,
-      "required_pages_per_second_upper_estimate": 0.0017365150784576133
+      "required_token_positions_per_second": 28877.496927965763,
+      "required_pages_per_second_upper_estimate": 0.0017311618474468284
     },
     {
       "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": 28966.79413163082,
-      "required_pages_per_second_upper_estimate": 0.00225204083227217
+      "required_token_positions_per_second": 28877.496927965763,
+      "required_pages_per_second_upper_estimate": 0.0022450983674641004
     },
     {
       "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": 28966.79413163082,
-      "required_pages_per_second_upper_estimate": 0.37155493300022857
+      "required_token_positions_per_second": 28877.496927965763,
+      "required_pages_per_second_upper_estimate": 0.3704095243514804
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.09559450175004258,
-    "producer_p95_seconds": 0.2698855479975464,
-    "gpu_update_seconds": 0.9544722096053171,
-    "producer_batches_per_second": 10.460852681828582,
-    "gpu_batches_per_second": 1.0476994405248414,
-    "producer_headroom": 9.984593173580633,
+    "producer_batch_seconds": 0.49885689808601796,
+    "producer_p95_seconds": 0.2705506839993177,
+    "gpu_update_seconds": 0.9574237015405901,
+    "producer_batches_per_second": 2.0045828850653074,
+    "gpu_batches_per_second": 1.044469651619132,
+    "producer_headroom": 1.919235165864142,
     "sustainable": true,
-    "buffer_seconds": 60.13174920513498,
+    "buffer_seconds": 60.317693197057174,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 437612.0674872266,
+    "generation_bytes_per_second": 377860.2996030521,
     "upload_bytes_per_second": 5973939.174584315,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.90136259520148,
+    "estimated_full_shards_per_hour": 81.08002415722534,
     "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": [

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -318,3 +318,4 @@
 {"step": 556800, "loss": 0.5332759022712708, "lr": 0.0003, "grad_norm": 0.16874244225295668, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81497664318, "tokens_per_second": 153271.84258360954, "data_wait_s": 9.954001143341884e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 556900, "loss": 0.640224039554596, "lr": 0.0003, "grad_norm": 0.20323349065555812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81512380496, "tokens_per_second": 153542.17013608004, "data_wait_s": 1.0444000508869067e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 557000, "loss": 0.629692792892456, "lr": 0.0003, "grad_norm": 0.18588088466874614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81527117890, "tokens_per_second": 153583.02355411623, "data_wait_s": 1.0056002793135121e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 557100, "loss": 0.6559820771217346, "lr": 0.0003, "grad_norm": 0.19969871289414715, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81541656022, "tokens_per_second": 151686.24923144432, "data_wait_s": 9.815998055273667e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7d844c7eb90b8ef181a8ad5f79752cf61effcf94172e4e3b21b6fd5132e95e20
-size 682623
+oid sha256:5877c0ea356bdd4a6b5693cd420de647c4cfb414bec9a2fe1d0b61721edace8c
+size 684741