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": 75743.07913097787,
-      "required_pages_per_second_upper_estimate": 0.005240081962438004
+      "required_token_positions_per_second": 75145.97078445953,
+      "required_pages_per_second_upper_estimate": 0.005198772621543619
     },
     {
       "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": 9467.884891372234,
-      "required_pages_per_second_upper_estimate": 0.003964005692098451
+      "required_token_positions_per_second": 9393.24634805744,
+      "required_pages_per_second_upper_estimate": 0.003932756092642569
     },
     {
       "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": 18935.769782744468,
-      "required_pages_per_second_upper_estimate": 0.003943821475915335
+      "required_token_positions_per_second": 18786.49269611488,
+      "required_pages_per_second_upper_estimate": 0.003912730995471898
     },
     {
       "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": 28403.6546741167,
-      "required_pages_per_second_upper_estimate": 0.0017027557278438056
+      "required_token_positions_per_second": 28179.739044172322,
+      "required_pages_per_second_upper_estimate": 0.0016893323277280105
     },
     {
       "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": 28403.6546741167,
-      "required_pages_per_second_upper_estimate": 0.002208259216439144
+      "required_token_positions_per_second": 28179.739044172322,
+      "required_pages_per_second_upper_estimate": 0.0021908507611117373
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28403.6546741167,
-      "required_pages_per_second_upper_estimate": 0.3643315846912777
+      "required_token_positions_per_second": 28179.739044172322,
+      "required_pages_per_second_upper_estimate": 0.3614594354122231
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08605613933589495,
-    "producer_p95_seconds": 0.22428638800010958,
-    "gpu_update_seconds": 0.9733958646242343,
-    "producer_batches_per_second": 11.620321428745399,
-    "gpu_batches_per_second": 1.027331259914522,
-    "producer_headroom": 11.311172824345146,
+    "producer_batch_seconds": 0.31968721835163194,
+    "producer_p95_seconds": 0.2393177169997216,
+    "gpu_update_seconds": 0.9811304482508227,
+    "producer_batches_per_second": 3.1280574968126347,
+    "gpu_batches_per_second": 1.0192324596416493,
+    "producer_headroom": 3.069032454002127,
     "sustainable": true,
-    "buffer_seconds": 60.35054360670253,
+    "buffer_seconds": 60.83008779155101,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 551112.5693575254,
-    "upload_bytes_per_second": 3495788.8247184367,
+    "generation_bytes_per_second": 478963.6825099134,
+    "upload_bytes_per_second": 3481450.2812785017,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 118.25592814010928,
+    "estimated_full_shards_per_hour": 102.7744565627389,
     "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": 1.3218913410000823,
-      "bytes": 7396091,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.1367835250002827,
       "bytes": 5040157,
@@ -562,6 +556,12 @@
       "bytes": 7422482,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.451942487999986,
+      "bytes": 6077412,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -155,3 +155,4 @@
 {"step": 858000, "loss": 0.49318116903305054, "lr": 3e-05, "grad_norm": 0.31133617668927394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125587339788, "tokens_per_second": 153133.82451254097, "data_wait_s": 1.4123999790172093e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 858100, "loss": 0.5367608666419983, "lr": 3e-05, "grad_norm": 0.30395223893279827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125602043141, "tokens_per_second": 153264.8200701733, "data_wait_s": 1.3453998690238222e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 858200, "loss": 0.4901749789714813, "lr": 3e-05, "grad_norm": 0.3019906820753487, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125616557291, "tokens_per_second": 150899.9502123738, "data_wait_s": 1.3232000128482468e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 858300, "loss": 0.5474139451980591, "lr": 3e-05, "grad_norm": 0.297106369941837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125631118251, "tokens_per_second": 150321.89091603263, "data_wait_s": 1.5887000699876808e-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:daa2749fd4b558435bfbed5dfa37cac7d5962553a17d3c35501230ce28bf230e
-size 335040
+oid sha256:27e38cc29b0f4420bb1aee0df099221001eb6b757041ee3b3e0b3831c25f4500
+size 337158