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": 75197.55351305181,
-      "required_pages_per_second_upper_estimate": 0.005202341234396053
+      "required_token_positions_per_second": 76847.4465022546,
+      "required_pages_per_second_upper_estimate": 0.005316484659668271
     },
     {
       "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": 9399.694189131476,
-      "required_pages_per_second_upper_estimate": 0.003935455669054035
+      "required_token_positions_per_second": 9605.930812781826,
+      "required_pages_per_second_upper_estimate": 0.004021802636666002
     },
     {
       "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": 18799.388378262953,
-      "required_pages_per_second_upper_estimate": 0.003915416825981334
+      "required_token_positions_per_second": 19211.86162556365,
+      "required_pages_per_second_upper_estimate": 0.004001324125743048
     },
     {
       "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": 28199.08256739443,
-      "required_pages_per_second_upper_estimate": 0.0016904919424093263
+      "required_token_positions_per_second": 28817.792438345474,
+      "required_pages_per_second_upper_estimate": 0.0017275826544575692
     },
     {
       "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": 28199.08256739443,
-      "required_pages_per_second_upper_estimate": 0.0021923546349590897
+      "required_token_positions_per_second": 28817.792438345474,
+      "required_pages_per_second_upper_estimate": 0.002240456606008419
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28199.08256739443,
-      "required_pages_per_second_upper_estimate": 0.36170755335865923
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28817.792438345474,
+      "required_pages_per_second_upper_estimate": 0.3696436992643177
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08594555653888847,
-    "producer_p95_seconds": 0.29909132800094085,
-    "gpu_update_seconds": 0.9804574292061676,
-    "producer_batches_per_second": 11.635272843308917,
-    "gpu_batches_per_second": 1.0199320951748565,
-    "producer_headroom": 11.407889700062997,
+    "producer_batch_seconds": 0.13426082494521552,
+    "producer_p95_seconds": 0.30513701499876333,
+    "gpu_update_seconds": 0.959407284897058,
+    "producer_batches_per_second": 7.448189003814368,
+    "gpu_batches_per_second": 1.0423102010396945,
+    "producer_headroom": 7.145846789549665,
     "sustainable": true,
-    "buffer_seconds": 60.78836061078239,
+    "buffer_seconds": 60.44265894851465,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348936.2302639156,
-    "upload_bytes_per_second": 5192858.902495448,
+    "generation_bytes_per_second": 363761.0589169544,
+    "upload_bytes_per_second": 5183803.302839173,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.87359219492055,
+    "estimated_full_shards_per_hour": 78.05465531951403,
     "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": 2.282538455998292,
-      "bytes": 6276829,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.4166184950008756,
       "bytes": 8285024,
@@ -562,6 +556,12 @@
       "bytes": 10517926,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.3904311470032553,
+      "bytes": 899587,
+      "pages": 31,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -360,3 +360,4 @@
 {"step": 756000, "loss": 0.545495867729187, "lr": 5e-05, "grad_norm": 0.2460345661702025, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110657106464, "tokens_per_second": 148597.34362736263, "data_wait_s": 8.72999953571707e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 756100, "loss": 0.4801879823207855, "lr": 5e-05, "grad_norm": 0.24014653203349337, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110671700511, "tokens_per_second": 150844.2225975514, "data_wait_s": 9.760995453689247e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 756200, "loss": 0.5434004068374634, "lr": 5e-05, "grad_norm": 0.25498685783512454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110686359307, "tokens_per_second": 151903.76877241887, "data_wait_s": 1.4113997167441994e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 756300, "loss": 0.5599837303161621, "lr": 5e-05, "grad_norm": 0.24912347089497042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110701043673, "tokens_per_second": 152258.57006895452, "data_wait_s": 1.3857999874744564e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9b4990a47606be46e1ec55dda49581d7b66b0412075e647982d2c461e71c0df9
-size 772362
+oid sha256:4905692ab57381af2974f6e2ef8c2defa80261610478afb2387440fdb229b760
+size 774480