brucethemoose
commited on
Commit
•
6873837
1
Parent(s):
add2ac4
Update README.md
Browse files
README.md
CHANGED
@@ -51,6 +51,68 @@ I am a huge fan of Kalomaze's quadratic sampling (shown as "smoothing factor" wh
|
|
51 |
|
52 |
Otherwise, I recommend a lower temperature with 0.1 or higher MinP, a little repetition penalty, and mirostat with a low tau, and no other samplers. See the explanation here: https://github.com/ggerganov/llama.cpp/pull/3841
|
53 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
54 |
24GB GPUs can efficiently run Yi-34B-200K models at **40K-90K context** with exllamav2, and performant UIs like [exui](https://github.com/turboderp/exui). I go into more detail in this [post](https://old.reddit.com/r/LocalLLaMA/comments/1896igc/how_i_run_34b_models_at_75k_context_on_24gb_fast/). Empty 16GB GPUs can still run the high context with aggressive quantization.
|
55 |
|
56 |
To load/train this in full-context backends like transformers, you *must* change `max_position_embeddings` in config.json to a lower value than 200,000, otherwise you will OOM! I do not recommend running high context without context-efficient backends that support flash attention + 8 bit kv cache, like exllamav2, litellm, vllm or unsloth.
|
|
|
51 |
|
52 |
Otherwise, I recommend a lower temperature with 0.1 or higher MinP, a little repetition penalty, and mirostat with a low tau, and no other samplers. See the explanation here: https://github.com/ggerganov/llama.cpp/pull/3841
|
53 |
|
54 |
+
@MarinaraSpaghetti has extensively tested the model and recommended the following settings, and they seem to work quite well:
|
55 |
+
|
56 |
+
```
|
57 |
+
{
|
58 |
+
"temp": 1,
|
59 |
+
"temperature_last": true,
|
60 |
+
"top_p": 1,
|
61 |
+
"top_k": 0,
|
62 |
+
"top_a": 0,
|
63 |
+
"tfs": 1,
|
64 |
+
"epsilon_cutoff": 0,
|
65 |
+
"eta_cutoff": 0,
|
66 |
+
"typical_p": 0.9,
|
67 |
+
"min_p": 0,
|
68 |
+
"rep_pen": 1.1,
|
69 |
+
"rep_pen_range": 19456,
|
70 |
+
"no_repeat_ngram_size": 0,
|
71 |
+
"penalty_alpha": 0,
|
72 |
+
"num_beams": 1,
|
73 |
+
"length_penalty": 0,
|
74 |
+
"min_length": 0,
|
75 |
+
"encoder_rep_pen": 1,
|
76 |
+
"freq_pen": 0,
|
77 |
+
"presence_pen": 0,
|
78 |
+
"do_sample": true,
|
79 |
+
"early_stopping": false,
|
80 |
+
"dynatemp": false,
|
81 |
+
"min_temp": 1,
|
82 |
+
"max_temp": 2,
|
83 |
+
"dynatemp_exponent": 1,
|
84 |
+
"smoothing_factor": 0.33,
|
85 |
+
"add_bos_token": false,
|
86 |
+
"truncation_length": 2048,
|
87 |
+
"ban_eos_token": false,
|
88 |
+
"skip_special_tokens": true,
|
89 |
+
"streaming": true,
|
90 |
+
"mirostat_mode": 0,
|
91 |
+
"mirostat_tau": 5,
|
92 |
+
"mirostat_eta": 0.1,
|
93 |
+
"guidance_scale": 1,
|
94 |
+
"negative_prompt": "",
|
95 |
+
"grammar_string": "",
|
96 |
+
"banned_tokens": "",
|
97 |
+
"ignore_eos_token_aphrodite": false,
|
98 |
+
"spaces_between_special_tokens_aphrodite": true,
|
99 |
+
"sampler_order": [
|
100 |
+
6,
|
101 |
+
0,
|
102 |
+
1,
|
103 |
+
3,
|
104 |
+
4,
|
105 |
+
2,
|
106 |
+
5
|
107 |
+
],
|
108 |
+
"logit_bias": [],
|
109 |
+
"n": 1,
|
110 |
+
"rep_pen_size": 0,
|
111 |
+
"genamt": 400,
|
112 |
+
"max_length": 38912
|
113 |
+
}
|
114 |
+
```
|
115 |
+
|
116 |
24GB GPUs can efficiently run Yi-34B-200K models at **40K-90K context** with exllamav2, and performant UIs like [exui](https://github.com/turboderp/exui). I go into more detail in this [post](https://old.reddit.com/r/LocalLLaMA/comments/1896igc/how_i_run_34b_models_at_75k_context_on_24gb_fast/). Empty 16GB GPUs can still run the high context with aggressive quantization.
|
117 |
|
118 |
To load/train this in full-context backends like transformers, you *must* change `max_position_embeddings` in config.json to a lower value than 200,000, otherwise you will OOM! I do not recommend running high context without context-efficient backends that support flash attention + 8 bit kv cache, like exllamav2, litellm, vllm or unsloth.
|