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": 77276.8302098129,
-      "required_pages_per_second_upper_estimate": 0.005346190420864621
+      "required_token_positions_per_second": 77190.96985978271,
+      "required_pages_per_second_upper_estimate": 0.005340250402626072
     },
     {
       "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": 9659.603776226613,
-      "required_pages_per_second_upper_estimate": 0.004044274385641406
+      "required_token_positions_per_second": 9648.871232472839,
+      "required_pages_per_second_upper_estimate": 0.004039780893692699
     },
     {
       "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": 19319.207552453227,
-      "required_pages_per_second_upper_estimate": 0.004023681451411684
+      "required_token_positions_per_second": 19297.742464945677,
+      "required_pages_per_second_upper_estimate": 0.004019210839756286
     },
     {
       "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": 28978.81132867984,
-      "required_pages_per_second_upper_estimate": 0.0017372354910715046
+      "required_token_positions_per_second": 28946.613697418517,
+      "required_pages_per_second_upper_estimate": 0.0017353052922403247
     },
     {
       "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": 28978.81132867984,
-      "required_pages_per_second_upper_estimate": 0.0022529751164846673
+      "required_token_positions_per_second": 28946.613697418517,
+      "required_pages_per_second_upper_estimate": 0.002250471892276512
     },
     {
       "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": 28978.81132867984,
-      "required_pages_per_second_upper_estimate": 0.3717090767009125
+      "required_token_positions_per_second": 28946.613697418517,
+      "required_pages_per_second_upper_estimate": 0.37129608005821524
     }
   ],
   "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": 0.2931807382498164,
-    "producer_p95_seconds": 0.32172993600033806,
-    "gpu_update_seconds": 0.95407640038835,
-    "producer_batches_per_second": 3.4108652770630177,
-    "gpu_batches_per_second": 1.048134090302367,
-    "producer_headroom": 3.254226065749896,
+    "producer_batch_seconds": 0.08820490336725584,
+    "producer_p95_seconds": 0.3352806110015081,
+    "gpu_update_seconds": 0.9551376298798526,
+    "producer_batches_per_second": 11.337238201331427,
+    "gpu_batches_per_second": 1.0469695347735286,
+    "producer_headroom": 10.828622825003023,
     "sustainable": true,
-    "buffer_seconds": 60.106813224466045,
+    "buffer_seconds": 60.17367068243072,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 156,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 556542.5542228261,
-    "upload_bytes_per_second": 4838217.725738191,
+    "generation_bytes_per_second": 559609.5699144317,
+    "upload_bytes_per_second": 4795149.072610317,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 119.42107648862446,
+    "estimated_full_shards_per_hour": 60.03959332978588,
     "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": 1.2196332230014377,
-      "bytes": 5954986,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.0563156829994114,
       "bytes": 932363,
@@ -562,6 +556,12 @@
       "bytes": 15617223,
       "pages": 32,
       "shards": 4
+    },
+    {
+      "seconds": 1.0353363290014386,
+      "bytes": 1240846,
+      "pages": 36,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -327,3 +327,4 @@
 {"step": 957700, "loss": 0.6012611389160156, "lr": 3e-05, "grad_norm": 0.3550671449172152, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140181537401, "tokens_per_second": 154312.0595150318, "data_wait_s": 1.1293002899037674e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 957800, "loss": 0.5727612972259521, "lr": 3e-05, "grad_norm": 0.3526797761891248, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140196128522, "tokens_per_second": 151676.82475969457, "data_wait_s": 1.4035998901817948e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 957900, "loss": 0.506212055683136, "lr": 3e-05, "grad_norm": 0.3587087317835405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140210585668, "tokens_per_second": 150557.25984268147, "data_wait_s": 1.3520002539735287e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 958000, "loss": 0.5144978761672974, "lr": 3e-05, "grad_norm": 0.3396069652641457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140225274890, "tokens_per_second": 152352.72301761995, "data_wait_s": 1.3647000741912052e-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 sha256:53309d68772bd6c751b76a5616537b58c8a2765196b2a47353bccbe18362e2a9
-size 701685
+oid sha256:8fb5195ad8d57781e1cd0e5b167041c57b39fb912a7e65c96bd4748fb150f21e
+size 703803