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": 76965.08483078974,
-      "required_pages_per_second_upper_estimate": 0.0053246231521948606
+      "required_token_positions_per_second": 77478.81688760198,
+      "required_pages_per_second_upper_estimate": 0.005360164327907731
     },
     {
       "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": 9620.635603848717,
-      "required_pages_per_second_upper_estimate": 0.004027959225614754
+      "required_token_positions_per_second": 9684.852110950247,
+      "required_pages_per_second_upper_estimate": 0.004054845336145006
     },
     {
       "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": 19241.271207697435,
-      "required_pages_per_second_upper_estimate": 0.0040074493661186785
+      "required_token_positions_per_second": 19369.704221900494,
+      "required_pages_per_second_upper_estimate": 0.004034198575970825
     },
     {
       "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": 28861.90681154615,
-      "required_pages_per_second_upper_estimate": 0.0017302272437722018
+      "required_token_positions_per_second": 29054.55633285074,
+      "required_pages_per_second_upper_estimate": 0.0017417762884156256
     },
     {
       "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": 28861.90681154615,
-      "required_pages_per_second_upper_estimate": 0.0022438863044861503
+      "required_token_positions_per_second": 29054.55633285074,
+      "required_pages_per_second_upper_estimate": 0.0022588639573918917
     },
     {
       "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": 28861.90681154615,
-      "required_pages_per_second_upper_estimate": 0.3702095510774124
+      "required_token_positions_per_second": 29054.55633285074,
+      "required_pages_per_second_upper_estimate": 0.3726806522857678
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.058269827085965176,
-    "producer_p95_seconds": 0.18524640199939313,
-    "gpu_update_seconds": 0.957940865810691,
-    "producer_batches_per_second": 17.161540543525298,
-    "gpu_batches_per_second": 1.043905772987057,
-    "producer_headroom": 16.4397410069099,
+    "producer_batch_seconds": 0.10698524057808356,
+    "producer_p95_seconds": 0.20453298400025233,
+    "gpu_update_seconds": 0.9515891305743186,
+    "producer_batches_per_second": 9.347083715441537,
+    "gpu_batches_per_second": 1.050873709955539,
+    "producer_headroom": 8.894583266182384,
     "sustainable": true,
-    "buffer_seconds": 60.350274546073535,
+    "buffer_seconds": 60.90170435675639,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 362770.16591410193,
-    "upload_bytes_per_second": 5950355.542605752,
+    "generation_bytes_per_second": 378268.98111606756,
+    "upload_bytes_per_second": 5854391.437050558,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.84203274791044,
+    "estimated_full_shards_per_hour": 81.16771769629975,
     "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": [
@@ -460,6 +460,12 @@
       "bytes": 6120696,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.070229210999969,
+      "bytes": 1151046,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -61,3 +61,4 @@
 {"step": 491100, "loss": 0.625411868095398, "lr": 0.0003, "grad_norm": 0.19665559218478687, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71880471324, "tokens_per_second": 151413.50245062984, "data_wait_s": 1.2169000910944305e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 491200, "loss": 0.5591356754302979, "lr": 0.0003, "grad_norm": 0.14437696167595815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71894998167, "tokens_per_second": 151455.14696415327, "data_wait_s": 1.0092000593431294e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 491300, "loss": 0.5596840381622314, "lr": 0.0003, "grad_norm": 0.16264493569802665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71909687015, "tokens_per_second": 153820.41107018662, "data_wait_s": 8.785000318312086e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 491400, "loss": 0.5643754005432129, "lr": 0.0003, "grad_norm": 0.16921761438386776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71924314684, "tokens_per_second": 139882.67433042926, "data_wait_s": 1.0732999726315029e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f424982041a91637f91daca28522cc31d09edfec0d07996cb229475bbad4acf1
-size 134382
+oid sha256:7c7b8c4a11101bca1bd694832061722cbee64c882077995f86e668d36e0f4b51
+size 136500