Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 72792.23920646422,
-      "required_pages_per_second_upper_estimate": 0.006309446804172741
+      "required_token_positions_per_second": 75440.40070112812,
+      "required_pages_per_second_upper_estimate": 0.006538982730826262
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9099.029900808027,
-      "required_pages_per_second_upper_estimate": 0.0036297390700526675
+      "required_token_positions_per_second": 9430.050087641015,
+      "required_pages_per_second_upper_estimate": 0.003761787971773183
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18198.059801616055,
-      "required_pages_per_second_upper_estimate": 0.004209800860190112
+      "required_token_positions_per_second": 18860.10017528203,
+      "required_pages_per_second_upper_estimate": 0.0043629522491251075
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27297.089702424084,
-      "required_pages_per_second_upper_estimate": 0.0016622495461163327
+      "required_token_positions_per_second": 28290.150262923045,
+      "required_pages_per_second_upper_estimate": 0.001722721724064623
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27297.089702424084,
-      "required_pages_per_second_upper_estimate": 0.002101405528473161
+      "required_token_positions_per_second": 28290.150262923045,
+      "required_pages_per_second_upper_estimate": 0.0021778540793879445
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27297.089702424084,
-      "required_pages_per_second_upper_estimate": 0.3389384966217277
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28290.150262923045,
+      "required_pages_per_second_upper_estimate": 0.35126898522322714
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 60,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10635090525016722,
-    "producer_p95_seconds": 0.41679349500191165,
-    "gpu_update_seconds": 1.012855227476677,
-    "producer_batches_per_second": 9.402834866781049,
-    "gpu_batches_per_second": 0.9873079319453155,
-    "producer_headroom": 9.523710447919148,
+    "producer_batch_seconds": 0.5079399767184327,
+    "producer_p95_seconds": 0.44360478099406464,
+    "gpu_update_seconds": 0.9773012777608098,
+    "producer_batches_per_second": 1.968736555174376,
+    "gpu_batches_per_second": 1.0232259209679921,
+    "producer_headroom": 1.9240487509463327,
     "sustainable": true,
-    "buffer_seconds": 60.77131364860061,
+    "buffer_seconds": 60.59267922117021,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 293818.93534164334,
+    "generation_bytes_per_second": 265015.39008857415,
     "upload_bytes_per_second": 5599692.86246779,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.04670376955963,
+    "estimated_full_shards_per_hour": 56.866133470467744,
     "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": [

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -399,3 +399,4 @@
 {"step": 234900, "loss": 0.5683729182928801, "lr": 0.0003, "grad_norm": 0.15020179548666457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34382846136, "tokens_per_second": 146679.03862138622, "data_wait_s": 2.205999771831557e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 235000, "loss": 0.5422572437673807, "lr": 0.0003, "grad_norm": 0.15907714641820941, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34397487726, "tokens_per_second": 148088.95694078197, "data_wait_s": 1.057400368154049e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 235100, "loss": 0.6112591940909624, "lr": 0.0003, "grad_norm": 0.14457853247116872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34412172596, "tokens_per_second": 123930.859132861, "data_wait_s": 1.5904995962046087e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 235200, "loss": 0.5915999626740813, "lr": 0.0003, "grad_norm": 0.139044980268077, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34426840209, "tokens_per_second": 148063.79049555262, "data_wait_s": 1.612499909242615e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:94ed39f655e12a1be86ab539b7861c72f7ed38d28b60ae343651387cedf5ba12
-size 855747
+oid sha256:ff3a285a8aa8114b4c2d3d3538caa68cf20d9add075e1fbdc95018b3435ebead
+size 857865