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": 76255.50143964749,
-      "required_pages_per_second_upper_estimate": 0.005275532526735334
+      "required_token_positions_per_second": 75832.52282853934,
+      "required_pages_per_second_upper_estimate": 0.005246269884973277
     },
     {
       "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": 9531.937679955936,
-      "required_pages_per_second_upper_estimate": 0.00399082325710412
+      "required_token_positions_per_second": 9479.065353567417,
+      "required_pages_per_second_upper_estimate": 0.003968686718144974
     },
     {
       "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": 19063.87535991187,
-      "required_pages_per_second_upper_estimate": 0.00397050248926805
+      "required_token_positions_per_second": 18958.130707134835,
+      "required_pages_per_second_upper_estimate": 0.003948478666768617
     },
     {
       "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": 28595.813039867808,
-      "required_pages_per_second_upper_estimate": 0.001714275328461749
+      "required_token_positions_per_second": 28437.19606070225,
+      "required_pages_per_second_upper_estimate": 0.0017047664827548774
     },
     {
       "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": 28595.813039867808,
-      "required_pages_per_second_upper_estimate": 0.002223198684160967
+      "required_token_positions_per_second": 28437.19606070225,
+      "required_pages_per_second_upper_estimate": 0.002210866911713203
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28595.813039867808,
-      "required_pages_per_second_upper_estimate": 0.3667963858835547
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28437.19606070225,
+      "required_pages_per_second_upper_estimate": 0.3647618175844621
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06287530977336075,
-    "producer_p95_seconds": 0.2078481529988494,
-    "gpu_update_seconds": 0.966854831560607,
-    "producer_batches_per_second": 15.904494206145188,
-    "gpu_batches_per_second": 1.0342814322868854,
-    "producer_headroom": 15.377337066739155,
+    "producer_batch_seconds": 0.10770820156248817,
+    "producer_p95_seconds": 0.2197813259990653,
+    "gpu_update_seconds": 0.9722477539973482,
+    "producer_batches_per_second": 9.284344047095043,
+    "gpu_batches_per_second": 1.0285444177047978,
+    "producer_headroom": 9.026682647126805,
     "sustainable": true,
-    "buffer_seconds": 60.91185438831824,
+    "buffer_seconds": 60.27936074783559,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 325488.75312133785,
-    "upload_bytes_per_second": 5482064.587464745,
+    "generation_bytes_per_second": 325228.94946321944,
+    "upload_bytes_per_second": 5402316.224566561,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.84230942945577,
+    "estimated_full_shards_per_hour": 69.78656161234319,
     "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": [
@@ -442,6 +442,12 @@
       "bytes": 6070484,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9431054610013234,
+      "bytes": 842654,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -89,3 +89,4 @@
 {"step": 768900, "loss": 0.5399148464202881, "lr": 3e-05, "grad_norm": 0.24447142968711158, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112544190672, "tokens_per_second": 151908.01766839164, "data_wait_s": 1.2266000339877792e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 769000, "loss": 0.5726956129074097, "lr": 3e-05, "grad_norm": 0.24596094504859978, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112558881027, "tokens_per_second": 152140.10119972232, "data_wait_s": 1.1895999705302529e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 769100, "loss": 0.5401884317398071, "lr": 3e-05, "grad_norm": 0.24275489347306103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112573549843, "tokens_per_second": 152211.04967640928, "data_wait_s": 8.38999949337449e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 769200, "loss": 0.49378257989883423, "lr": 3e-05, "grad_norm": 0.26130389132820514, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112588212168, "tokens_per_second": 152889.41558694514, "data_wait_s": 1.3338998542167246e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:de2f5d6e2026f3914243d32d108c8be1502907085edb1ed714454ee26a78d04f
-size 193686
+oid sha256:326ddf275bd34bda804b7f51191609a4de4cdb0d38c97cfbc84e21d881811660
+size 195804