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": 76940.74094243823,
-      "required_pages_per_second_upper_estimate": 0.0053229389855131
+      "required_token_positions_per_second": 75331.62050999419,
+      "required_pages_per_second_upper_estimate": 0.0052116163003228175
     },
     {
       "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": 9617.59261780478,
-      "required_pages_per_second_upper_estimate": 0.004026685190902932
+      "required_token_positions_per_second": 9416.452563749273,
+      "required_pages_per_second_upper_estimate": 0.00394247205055185
     },
     {
       "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": 19235.18523560956,
-      "required_pages_per_second_upper_estimate": 0.00400618181863065
+      "required_token_positions_per_second": 18832.905127498547,
+      "required_pages_per_second_upper_estimate": 0.003922397480950991
     },
     {
       "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": 28852.77785341434,
-      "required_pages_per_second_upper_estimate": 0.0017296799766713118
+      "required_token_positions_per_second": 28249.35769124782,
+      "required_pages_per_second_upper_estimate": 0.0016935058593186675
     },
     {
       "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": 28852.77785341434,
-      "required_pages_per_second_upper_estimate": 0.0022431765681454456
+      "required_token_positions_per_second": 28249.35769124782,
+      "required_pages_per_second_upper_estimate": 0.002196263304701791
     },
     {
       "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": 28852.77785341434,
-      "required_pages_per_second_upper_estimate": 0.37009245460440915
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28249.35769124782,
+      "required_pages_per_second_upper_estimate": 0.36235242866622824
     }
   ],
   "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": 0.10345223277312243,
-    "producer_p95_seconds": 0.2940831810046802,
-    "gpu_update_seconds": 0.9582439562826437,
-    "producer_batches_per_second": 9.66629692945406,
-    "gpu_batches_per_second": 1.0435755878694422,
-    "producer_headroom": 9.262670612282829,
+    "producer_batch_seconds": 0.3912628681718502,
+    "producer_p95_seconds": 0.29599049799435306,
+    "gpu_update_seconds": 0.9787125180749107,
+    "producer_batches_per_second": 2.5558264822635324,
+    "gpu_batches_per_second": 1.0217504951984888,
+    "producer_headroom": 2.501419372218683,
     "sustainable": true,
-    "buffer_seconds": 60.36936924580655,
+    "buffer_seconds": 60.68017612064446,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 393043.2536075044,
-    "upload_bytes_per_second": 5031729.505582359,
+    "generation_bytes_per_second": 350595.7227495347,
+    "upload_bytes_per_second": 5204174.093604321,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.33793264550066,
+    "estimated_full_shards_per_hour": 75.22968065132648,
     "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": 2.882830210000975,
-      "bytes": 1047147,
-      "pages": 31,
-      "shards": 1
-    },
     {
       "seconds": 1.6690692930023943,
       "bytes": 11069831,
@@ -562,6 +556,12 @@
       "bytes": 6031838,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.433015736998641,
+      "bytes": 8073393,
+      "pages": 27,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -357,3 +357,4 @@
 {"step": 755700, "loss": 0.5509876012802124, "lr": 5e-05, "grad_norm": 0.24811867638996113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110613679862, "tokens_per_second": 151889.6556943709, "data_wait_s": 1.3719996786676347e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 755800, "loss": 0.5624434351921082, "lr": 5e-05, "grad_norm": 0.25050675376137443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110628188586, "tokens_per_second": 150407.52231608026, "data_wait_s": 9.04199987417087e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 755900, "loss": 0.5307029485702515, "lr": 5e-05, "grad_norm": 0.2428281234946559, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110642701618, "tokens_per_second": 145344.94013792335, "data_wait_s": 1.5197998436633497e-05, "gpu_peak_gb": 89.37899398803711}
+{"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}

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 sha256:207901c92508b9b1f61f205a289536ac2bebf09bd508df33d20179a54e01afc5
-size 766008
+oid sha256:cb3a02f86be16b137c958e4a82e11e302c7a82ae089fab147baa1d5dd266b932
+size 768126