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": 75331.62050999419,
-      "required_pages_per_second_upper_estimate": 0.0052116163003228175
+      "required_token_positions_per_second": 75700.54341147232,
+      "required_pages_per_second_upper_estimate": 0.005237139242666142
     },
     {
       "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": 9416.452563749273,
-      "required_pages_per_second_upper_estimate": 0.00394247205055185
+      "required_token_positions_per_second": 9462.56792643404,
+      "required_pages_per_second_upper_estimate": 0.003961779589909682
     },
     {
       "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": 18832.905127498547,
-      "required_pages_per_second_upper_estimate": 0.003922397480950991
+      "required_token_positions_per_second": 18925.13585286808,
+      "required_pages_per_second_upper_estimate": 0.003941606708757673
     },
     {
       "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": 28249.35769124782,
-      "required_pages_per_second_upper_estimate": 0.0016935058593186675
+      "required_token_positions_per_second": 28387.703779302123,
+      "required_pages_per_second_upper_estimate": 0.001701799496055278
     },
     {
       "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": 28249.35769124782,
-      "required_pages_per_second_upper_estimate": 0.002196263304701791
+      "required_token_positions_per_second": 28387.703779302123,
+      "required_pages_per_second_upper_estimate": 0.002207019104527882
     },
     {
       "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": 28249.35769124782,
-      "required_pages_per_second_upper_estimate": 0.36235242866622824
+      "required_token_positions_per_second": 28387.703779302123,
+      "required_pages_per_second_upper_estimate": 0.3641269837393328
     }
   ],
   "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": 0.3912628681718502,
-    "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,
+    "producer_batch_seconds": 0.10529721488308041,
+    "producer_p95_seconds": 0.2951533159939572,
+    "gpu_update_seconds": 0.9739428104135196,
+    "producer_batches_per_second": 9.496927350931141,
+    "gpu_batches_per_second": 1.0267543322953603,
+    "producer_headroom": 9.249464114458897,
     "sustainable": true,
-    "buffer_seconds": 60.68017612064446,
+    "buffer_seconds": 60.38445424563822,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 350595.7227495347,
-    "upload_bytes_per_second": 5204174.093604321,
+    "generation_bytes_per_second": 349294.5680909998,
+    "upload_bytes_per_second": 5192858.902495448,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.22968065132648,
+    "estimated_full_shards_per_hour": 74.95048315093513,
     "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": 1.6690692930023943,
-      "bytes": 11069831,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 2.282538455998292,
       "bytes": 6276829,
@@ -562,6 +556,12 @@
       "bytes": 8073393,
       "pages": 27,
       "shards": 3
+    },
+    {
+      "seconds": 1.7437503740002285,
+      "bytes": 10517926,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -358,3 +358,4 @@
 {"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}
+{"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}

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:cb3a02f86be16b137c958e4a82e11e302c7a82ae089fab147baa1d5dd266b932
-size 768126
+oid sha256:5a23d6466bbea6414f887a009093022817eaca83d2138e25d34b2dcad7af1917
+size 770244