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": 77623.52134861016,
-      "required_pages_per_second_upper_estimate": 0.005370175318280886
+      "required_token_positions_per_second": 77549.68345802694,
+      "required_pages_per_second_upper_estimate": 0.005365067041682828
     },
     {
       "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": 9702.94016857627,
-      "required_pages_per_second_upper_estimate": 0.004062418428151427
+      "required_token_positions_per_second": 9693.710432253367,
+      "required_pages_per_second_upper_estimate": 0.004058554130291839
     },
     {
       "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": 19405.88033715254,
-      "required_pages_per_second_upper_estimate": 0.004041733106749511
+      "required_token_positions_per_second": 19387.420864506734,
+      "required_pages_per_second_upper_estimate": 0.00403788848540642
     },
     {
       "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": 29108.82050572881,
-      "required_pages_per_second_upper_estimate": 0.0017450293427230712
+      "required_token_positions_per_second": 29081.1312967601,
+      "required_pages_per_second_upper_estimate": 0.0017433694169243712
     },
     {
       "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": 29108.82050572881,
-      "required_pages_per_second_upper_estimate": 0.00226308275815029
+      "required_token_positions_per_second": 29081.1312967601,
+      "required_pages_per_second_upper_estimate": 0.0022609300439449314
     },
     {
       "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": 29108.82050572881,
-      "required_pages_per_second_upper_estimate": 0.37337669483111824
+      "required_token_positions_per_second": 29081.1312967601,
+      "required_pages_per_second_upper_estimate": 0.3730215273888239
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05215023860159107,
-    "producer_p95_seconds": 0.18244040700028563,
-    "gpu_update_seconds": 0.9498151941456607,
-    "producer_batches_per_second": 19.175367684118143,
-    "gpu_batches_per_second": 1.0528363898194737,
-    "producer_headroom": 18.213055579705102,
+    "producer_batch_seconds": 0.4674273531797155,
+    "producer_p95_seconds": 0.20749890399929427,
+    "gpu_update_seconds": 0.95071954793864,
+    "producer_batches_per_second": 2.139369878971379,
+    "gpu_batches_per_second": 1.0518348993330477,
+    "producer_headroom": 2.033940764209212,
     "sustainable": true,
-    "buffer_seconds": 60.78817242532229,
+    "buffer_seconds": 60.84605106807296,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297605.9346875228,
-    "upload_bytes_per_second": 5256151.386890771,
+    "generation_bytes_per_second": 270112.40264893026,
+    "upload_bytes_per_second": 5183632.832951105,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.85930567235244,
+    "estimated_full_shards_per_hour": 57.95983371354037,
     "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": [
@@ -502,6 +502,12 @@
       "bytes": 6147791,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1267749469989212,
+      "bytes": 894977,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -79,3 +79,4 @@
 {"step": 892900, "loss": 0.5386309027671814, "lr": 3e-05, "grad_norm": 0.3150166992141552, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130695448725, "tokens_per_second": 154076.71778541556, "data_wait_s": 1.0897000720433425e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 893000, "loss": 0.5396263003349304, "lr": 3e-05, "grad_norm": 0.30559744526303734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130710170691, "tokens_per_second": 154258.12412871476, "data_wait_s": 1.2008000339847058e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 893100, "loss": 0.529788613319397, "lr": 3e-05, "grad_norm": 0.32204225317044316, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130724781340, "tokens_per_second": 135370.96086443754, "data_wait_s": 1.434399928257335e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 893200, "loss": 0.5528310537338257, "lr": 3e-05, "grad_norm": 0.3245383546079211, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130739254414, "tokens_per_second": 151356.05284035948, "data_wait_s": 1.2751999747706577e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f2d672599654b0374a0e056bc4dba67294cbb0d2c8389c7765e10fc25dd2c5b1
-size 172506
+oid sha256:51b5a6863b7d50de7bfe43fb5edb529cb1389fbbded2ced20b46b94126409f0e
+size 174624