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": 75928.47079212754,
-      "required_pages_per_second_upper_estimate": 0.005252907787724258
+      "required_token_positions_per_second": 77296.01927073179,
+      "required_pages_per_second_upper_estimate": 0.005347517964623744
     },
     {
       "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": 9491.058849015943,
-      "required_pages_per_second_upper_estimate": 0.0039737081442366064
+      "required_token_positions_per_second": 9662.002408841474,
+      "required_pages_per_second_upper_estimate": 0.004045278643028102
     },
     {
       "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": 18982.117698031885,
-      "required_pages_per_second_upper_estimate": 0.003953474524392983
+      "required_token_positions_per_second": 19324.004817682948,
+      "required_pages_per_second_upper_estimate": 0.004024680595246644
     },
     {
       "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": 28473.176547047828,
-      "required_pages_per_second_upper_estimate": 0.0017069234579722733
+      "required_token_positions_per_second": 28986.007226524423,
+      "required_pages_per_second_upper_estimate": 0.0017376668741597876
     },
     {
       "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": 28473.176547047828,
-      "required_pages_per_second_upper_estimate": 0.002213664236265138
+      "required_token_positions_per_second": 28986.007226524423,
+      "required_pages_per_second_upper_estimate": 0.002253534565890674
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28473.176547047828,
-      "required_pages_per_second_upper_estimate": 0.36522333663046685
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28986.007226524423,
+      "required_pages_per_second_upper_estimate": 0.3718013779521097
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10750935074196377,
-    "producer_p95_seconds": 0.3130095579981571,
-    "gpu_update_seconds": 0.9710191609396183,
-    "producer_batches_per_second": 9.301516501575088,
-    "gpu_batches_per_second": 1.0298457952491258,
-    "producer_headroom": 9.031950748825455,
+    "producer_batch_seconds": 0.19094193648433588,
+    "producer_p95_seconds": 0.32086455700118677,
+    "gpu_update_seconds": 0.9538395469211074,
+    "producer_batches_per_second": 5.2371941879935635,
+    "gpu_batches_per_second": 1.0483943585982503,
+    "producer_headroom": 4.9954429314136375,
     "sustainable": true,
-    "buffer_seconds": 60.20318797825633,
+    "buffer_seconds": 60.091891456029764,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 163,
+    "upload_pages_per_commit": 156,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 596519.4340587509,
-    "upload_bytes_per_second": 4865709.336489978,
+    "generation_bytes_per_second": 595572.2897738633,
+    "upload_bytes_per_second": 4872575.5915064365,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.999592143640015,
+    "estimated_full_shards_per_hour": 63.89797458606684,
     "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": 1.2455697520017566,
-      "bytes": 7910132,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.3195631069975207,
       "bytes": 10098822,
@@ -562,6 +556,12 @@
       "bytes": 8344893,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.776490742999158,
+      "bytes": 11106825,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -320,3 +320,4 @@
 {"step": 957000, "loss": 0.5587459206581116, "lr": 3e-05, "grad_norm": 0.3490230127132062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140078464846, "tokens_per_second": 152948.65334969288, "data_wait_s": 1.628300014999695e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 957100, "loss": 0.46292242407798767, "lr": 3e-05, "grad_norm": 0.3678985811875492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140093208546, "tokens_per_second": 153248.6847400724, "data_wait_s": 1.1796997569035739e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 957200, "loss": 0.5018696784973145, "lr": 3e-05, "grad_norm": 0.34455979385048385, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140107852297, "tokens_per_second": 152747.56742564187, "data_wait_s": 7.456001185346395e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 957300, "loss": 0.5067824125289917, "lr": 3e-05, "grad_norm": 0.33700110238978237, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140122488239, "tokens_per_second": 152071.51007782918, "data_wait_s": 1.7456000932725146e-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:f0572fd2f6cfbc5e875e7b8437f0a8a4b39e64d79b4b22d7a41a3f46d75e844e
-size 686859
+oid sha256:6f21632b323177e9da3545e11b15efb50cea413e5846fd58cd19a24e6ae48cd9
+size 688977