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": 77738.64360948627,
-      "required_pages_per_second_upper_estimate": 0.005378139743408742
+      "required_token_positions_per_second": 77115.59233839801,
+      "required_pages_per_second_upper_estimate": 0.005335035610796732
     },
     {
       "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": 9717.330451185784,
-      "required_pages_per_second_upper_estimate": 0.004068443338976759
+      "required_token_positions_per_second": 9639.449042299751,
+      "required_pages_per_second_upper_estimate": 0.004035836019424932
     },
     {
       "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": 19434.66090237157,
-      "required_pages_per_second_upper_estimate": 0.00404772733949009
+      "required_token_positions_per_second": 19278.898084599503,
+      "required_pages_per_second_upper_estimate": 0.004015286052289907
     },
     {
       "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": 29151.991353557354,
-      "required_pages_per_second_upper_estimate": 0.0017476173691323247
+      "required_token_positions_per_second": 28918.347126899254,
+      "required_pages_per_second_upper_estimate": 0.0017336107544982488
     },
     {
       "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": 29151.991353557354,
-      "required_pages_per_second_upper_estimate": 0.0022664391016804677
+      "required_token_positions_per_second": 28918.347126899254,
+      "required_pages_per_second_upper_estimate": 0.0022482742907501433
     },
     {
       "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": 29151.991353557354,
-      "required_pages_per_second_upper_estimate": 0.3739304441138179
+      "required_token_positions_per_second": 28918.347126899254,
+      "required_pages_per_second_upper_estimate": 0.37093350684187293
     }
   ],
   "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": 0.14697699227394878,
-    "producer_p95_seconds": 0.32218778499373,
-    "gpu_update_seconds": 0.9484086237774688,
-    "producer_batches_per_second": 6.803785983972996,
-    "gpu_batches_per_second": 1.0543978354151242,
-    "producer_headroom": 6.452769301536261,
+    "producer_batch_seconds": 0.09498383518757691,
+    "producer_p95_seconds": 0.3251124439993873,
+    "gpu_update_seconds": 0.9560712401256984,
+    "producer_batches_per_second": 10.52810720924429,
+    "gpu_batches_per_second": 1.0459471617078724,
+    "producer_headroom": 10.065620515718495,
     "sustainable": true,
-    "buffer_seconds": 60.698151921758004,
+    "buffer_seconds": 60.232488127919005,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 247630.47374639043,
-    "upload_bytes_per_second": 5107554.115370357,
+    "generation_bytes_per_second": 247610.3559792492,
+    "upload_bytes_per_second": 5059811.9670167295,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.13573512357507,
+    "estimated_full_shards_per_hour": 53.13141831906421,
     "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": 1.6951102359998913,
-      "bytes": 7495029,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.1257126100026653,
       "bytes": 946737,
@@ -562,6 +556,12 @@
       "bytes": 7471801,
       "pages": 42,
       "shards": 3
+    },
+    {
+      "seconds": 1.1862390570022399,
+      "bytes": 886866,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -402,3 +402,4 @@
 {"step": 925200, "loss": 0.5638466477394104, "lr": 3e-05, "grad_norm": 0.33840653702274737, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135422770685, "tokens_per_second": 154307.70573229526, "data_wait_s": 1.7597005353309214e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 925300, "loss": 0.5204133987426758, "lr": 3e-05, "grad_norm": 0.3171108751895482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135437541266, "tokens_per_second": 154289.62706155606, "data_wait_s": 1.407499803462997e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 925400, "loss": 0.5439279079437256, "lr": 3e-05, "grad_norm": 0.3367022718858882, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135452275707, "tokens_per_second": 153383.2879145928, "data_wait_s": 1.4245000784285367e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 925500, "loss": 0.462994247674942, "lr": 3e-05, "grad_norm": 0.33712264966163574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135466831988, "tokens_per_second": 140100.0005896525, "data_wait_s": 1.8733000615611672e-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:cde6be6dc314afffcd5d8de217a69d3c8c82487eeda0d57a1d3cf7d1b0e5e8a4
-size 862101
+oid sha256:7053fa409a68687491fee262edefc2195c95c80e6e36fcc9c73a7a099c962d96
+size 864219