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": 76985.66727282797,
-      "required_pages_per_second_upper_estimate": 0.00532604709329291
+      "required_token_positions_per_second": 76036.3450205836,
+      "required_pages_per_second_upper_estimate": 0.005260370777151542
     },
     {
       "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": 9623.208409103496,
-      "required_pages_per_second_upper_estimate": 0.004029036405448647
+      "required_token_positions_per_second": 9504.54312757295,
+      "required_pages_per_second_upper_estimate": 0.0039793537300846695
     },
     {
       "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": 19246.41681820699,
-      "required_pages_per_second_upper_estimate": 0.004008521061088945
+      "required_token_positions_per_second": 19009.0862551459,
+      "required_pages_per_second_upper_estimate": 0.003959091363631155
     },
     {
       "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": 28869.625227310487,
-      "required_pages_per_second_upper_estimate": 0.0017306899510119355
+      "required_token_positions_per_second": 28513.62938271885,
+      "required_pages_per_second_upper_estimate": 0.0017093485437028476
     },
     {
       "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": 28869.625227310487,
-      "required_pages_per_second_upper_estimate": 0.0022444863773621046
+      "required_token_positions_per_second": 28513.62938271885,
+      "required_pages_per_second_upper_estimate": 0.00221680926630534
     },
     {
       "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": 28869.625227310487,
-      "required_pages_per_second_upper_estimate": 0.37030855462744816
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28513.62938271885,
+      "required_pages_per_second_upper_estimate": 0.36574222217158386
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0892782621171051,
-    "producer_p95_seconds": 0.20011891499962076,
-    "gpu_update_seconds": 0.9576847562899834,
-    "producer_batches_per_second": 11.200934878059268,
-    "gpu_batches_per_second": 1.0441849402239036,
-    "producer_headroom": 10.726964588914164,
+    "producer_batch_seconds": 0.08071264542967072,
+    "producer_p95_seconds": 0.21324933399955626,
+    "gpu_update_seconds": 0.9696415573373666,
+    "producer_batches_per_second": 12.389632314447605,
+    "gpu_batches_per_second": 1.0313089331133842,
+    "producer_headroom": 12.013502372218335,
     "sustainable": true,
-    "buffer_seconds": 60.33413964626895,
+    "buffer_seconds": 60.11777655491673,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 361524.11266266124,
-    "upload_bytes_per_second": 4790553.406314459,
+    "generation_bytes_per_second": 361485.2228068636,
+    "upload_bytes_per_second": 4783637.260420177,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.5746587267864,
+    "estimated_full_shards_per_hour": 77.56631386904174,
     "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": 1.5191957560000446,
-      "bytes": 7567101,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.065770677000728,
       "bytes": 897206,
@@ -562,6 +556,12 @@
       "bytes": 18268501,
       "pages": 39,
       "shards": 5
+    },
+    {
+      "seconds": 1.3536671069996373,
+      "bytes": 6151717,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -132,3 +132,4 @@
 {"step": 855700, "loss": 0.508650541305542, "lr": 3e-05, "grad_norm": 0.3107761555130699, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125250457357, "tokens_per_second": 141593.71435859895, "data_wait_s": 7.3999999585794285e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 855800, "loss": 0.5231516361236572, "lr": 3e-05, "grad_norm": 0.2864965672270696, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125265107838, "tokens_per_second": 152522.0216723379, "data_wait_s": 1.245400017069187e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 855900, "loss": 0.526910662651062, "lr": 3e-05, "grad_norm": 0.30600752623208666, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125279875021, "tokens_per_second": 153814.19635047676, "data_wait_s": 1.024799894366879e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 856000, "loss": 0.5293590426445007, "lr": 3e-05, "grad_norm": 0.2959618808929612, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125294573760, "tokens_per_second": 153006.3369240086, "data_wait_s": 1.4912999176885933e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6be9f3be85f37e6abdb3331c11041fa08b2ea39200d9964f654a1f061a0d5cf8
-size 286326
+oid sha256:3d01dcf2ee612eb5a5aedfe25d20780a3af2425c63d8794bb75bfd62b505ee21
+size 288444