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": 75800.03482084384,
-      "required_pages_per_second_upper_estimate": 0.0065701549048388765
+      "required_token_positions_per_second": 74301.83833943604,
+      "required_pages_per_second_upper_estimate": 0.006440295030974716
     },
     {
       "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": 9475.00435260548,
-      "required_pages_per_second_upper_estimate": 0.003779720900193665
+      "required_token_positions_per_second": 9287.729792429505,
+      "required_pages_per_second_upper_estimate": 0.0037050142781352743
     },
     {
       "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": 18950.00870521096,
-      "required_pages_per_second_upper_estimate": 0.004383751005188083
+      "required_token_positions_per_second": 18575.45958485901,
+      "required_pages_per_second_upper_estimate": 0.004297105658034038
     },
     {
       "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": 28425.013057816435,
-      "required_pages_per_second_upper_estimate": 0.0017309341607032293
+      "required_token_positions_per_second": 27863.189377288512,
+      "required_pages_per_second_upper_estimate": 0.0016967220462201233
     },
     {
       "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": 28425.013057816435,
-      "required_pages_per_second_upper_estimate": 0.0021882361906629473
+      "required_token_positions_per_second": 27863.189377288512,
+      "required_pages_per_second_upper_estimate": 0.0021449854485084212
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28425.013057816435,
-      "required_pages_per_second_upper_estimate": 0.35294352977906346
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27863.189377288512,
+      "required_pages_per_second_upper_estimate": 0.34596755996980905
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.13014643728905106,
-    "producer_p95_seconds": 0.15005702999997084,
-    "gpu_update_seconds": 0.9726644608311703,
-    "producer_batches_per_second": 7.683652513507012,
-    "gpu_batches_per_second": 1.028103770899032,
-    "producer_headroom": 7.473615729264365,
+    "producer_batch_seconds": 0.08865531217186273,
+    "producer_p95_seconds": 0.15351809100002356,
+    "gpu_update_seconds": 0.9922769294507284,
+    "producer_batches_per_second": 11.279639939245268,
+    "gpu_batches_per_second": 1.0077831806021598,
+    "producer_headroom": 11.192526484224095,
     "sustainable": true,
-    "buffer_seconds": 60.30519657153256,
+    "buffer_seconds": 60.52889269649443,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 129,
+    "upload_pages_per_commit": 134,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 662967.2275332123,
-    "upload_bytes_per_second": 4294338.088484311,
+    "generation_bytes_per_second": 651047.7203473067,
+    "upload_bytes_per_second": 4663796.135280369,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.12866697071684,
+    "estimated_full_shards_per_hour": 69.84984258563233,
     "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": [
@@ -256,6 +256,12 @@
       "bytes": 5580567,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.1845111140000881,
+      "bytes": 11699180,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -14,3 +14,4 @@
 {"step": 238500, "loss": 0.5690376684069633, "lr": 0.0003, "grad_norm": 0.13665270180605898, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34909089152, "tokens_per_second": 148479.06344308285, "data_wait_s": 1.3035999927524244e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 238600, "loss": 0.581579502671957, "lr": 0.0003, "grad_norm": 0.14854825603106736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34923699189, "tokens_per_second": 148238.670523885, "data_wait_s": 1.4490000012301607e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 238700, "loss": 0.6140376031398773, "lr": 0.0003, "grad_norm": 0.15842745276466186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34938361022, "tokens_per_second": 149039.28227337412, "data_wait_s": 1.0464000297361054e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 238800, "loss": 0.6271719429641962, "lr": 0.0003, "grad_norm": 0.14616646232827346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34953098704, "tokens_per_second": 149672.86953364583, "data_wait_s": 1.6234000213444233e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:915f984a347aa92a04d86aad1e3aac294f927bf98a934416db1818e91ea861bf
-size 34053
+oid sha256:f8c880f352fd38d25e7bc6a4480de915bd3679dd8e5687cfa7098b88eb202111
+size 36171