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": 77572.96483563015,
-      "required_pages_per_second_upper_estimate": 0.005366677701405657
+      "required_token_positions_per_second": 76800.98678963768,
+      "required_pages_per_second_upper_estimate": 0.005313270468948047
     },
     {
       "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": 9696.620604453768,
-      "required_pages_per_second_upper_estimate": 0.004059772558620841
+      "required_token_positions_per_second": 9600.12334870471,
+      "required_pages_per_second_upper_estimate": 0.004019371172730554
     },
     {
       "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": 19393.241208907537,
-      "required_pages_per_second_upper_estimate": 0.004039100709652303
+      "required_token_positions_per_second": 19200.24669740942,
+      "required_pages_per_second_upper_estimate": 0.003998905042514775
     },
     {
       "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": 29089.861813361305,
-      "required_pages_per_second_upper_estimate": 0.0017438927980638864
+      "required_token_positions_per_second": 28800.370046114127,
+      "required_pages_per_second_upper_estimate": 0.0017265382086457525
     },
     {
       "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": 29089.861813361305,
-      "required_pages_per_second_upper_estimate": 0.002261608803209191
+      "required_token_positions_per_second": 28800.370046114127,
+      "required_pages_per_second_upper_estimate": 0.0022391020916454367
     },
     {
       "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": 29089.861813361305,
-      "required_pages_per_second_upper_estimate": 0.3731335130816858
+      "required_token_positions_per_second": 28800.370046114127,
+      "required_pages_per_second_upper_estimate": 0.3694202235234813
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.22078416554685987,
-    "producer_p95_seconds": 0.2989882099973329,
-    "gpu_update_seconds": 0.9504342157892602,
-    "producer_batches_per_second": 4.529310322246625,
-    "gpu_batches_per_second": 1.0521506732263204,
-    "producer_headroom": 4.304811504190672,
+    "producer_batch_seconds": 0.14814580309371195,
+    "producer_p95_seconds": 0.3027946710026299,
+    "gpu_update_seconds": 0.9599876652880156,
+    "producer_batches_per_second": 6.750106848234062,
+    "gpu_batches_per_second": 1.0416800508577158,
+    "producer_headroom": 6.480019313680863,
     "sustainable": true,
-    "buffer_seconds": 60.827789810512655,
+    "buffer_seconds": 60.47922291314498,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
     "upload_pages_per_commit": 166,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 582723.2971245584,
-    "upload_bytes_per_second": 4728856.921417192,
+    "generation_bytes_per_second": 572208.251266587,
+    "upload_bytes_per_second": 4797016.458320768,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.51942722941667,
+    "estimated_full_shards_per_hour": 61.391285197726276,
     "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": 2.108060000999103,
-      "bytes": 5732876,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.102688891998696,
       "bytes": 2347731,
@@ -562,6 +556,12 @@
       "bytes": 10736001,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.2456738490000134,
+      "bytes": 7708205,
+      "pages": 43,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -326,3 +326,4 @@
 {"step": 917600, "loss": 0.5721331834793091, "lr": 3e-05, "grad_norm": 0.3503418495413616, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134310001711, "tokens_per_second": 151304.23800890677, "data_wait_s": 1.428399991709739e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 917700, "loss": 0.5132466554641724, "lr": 3e-05, "grad_norm": 0.32683448475066307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134324520310, "tokens_per_second": 151784.09833597168, "data_wait_s": 1.4648001524619758e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 917800, "loss": 0.5411073565483093, "lr": 3e-05, "grad_norm": 0.33555589197663044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134339171440, "tokens_per_second": 152849.26160328538, "data_wait_s": 1.454599987482652e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 917900, "loss": 0.5300824046134949, "lr": 3e-05, "grad_norm": 0.33125594738717046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134353833516, "tokens_per_second": 152752.2748519518, "data_wait_s": 1.5485999028896913e-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:ba9ff263cc24f66c53cb2ffd283e53aa0f2a0e67a058bce399000bfff2a2e1e9
-size 699567
+oid sha256:9e4db6cf72ae8c8a35f27f97636da7fb4bf48e5d32d676d1253ac84a7b5145b6
+size 701685